# Testing agent-browser, Playwright and Chrome DevTools with Claude Code and Codex

Published 1 Nov 2026 · 4 min read

I hand a lot of browser jobs to coding agents, and I wanted to find out which tool is fastest and cheapest for those jobs. So I had Claude Code and Codex do the same job in a signed-in web app with each tool, over its CLI with and without its skill and over its MCP server. The tools were mostly close on speed and cost. The big exception was agent-browser, which with Claude Code used 2.4 times as many input tokens over MCP as over its CLI with the skill.

- 18 setups, 12 runs each
- macOS, WSL and Windows
- 19 to 22 October 2026
- [Download every run (CSV)](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run)

I stuck to open-source tools, because I think building production workflows around proprietary ones, like the built-in browser in OpenAI’s desktop app or the Claude in Chrome extension, is a bad long-term strategy. Those tie you to one vendor’s agents and models, while all three tools I tested ran with both Claude Code and Codex, and work with any other agent harness. For a production workflow, it’s also important to be able to study a tool’s code, pin a version and fork it if it doesn’t fit your needs or its maintainers take it in a new direction.

## Key results

1. Speed had no clear winner, with one exception. Chrome DevTools over MCP with Claude Code was clearly faster than the other two tools.
2. With Claude Code, MCP was the most expensive connection for every tool.
3. agent-browser was cheaper over its CLI without its skill, with both Claude Code and Codex.
4. A Python script did the same job faster than every agent run.
5. Windows was the slowest place to run, slower than WSL on the same PC.

**With Claude Code, every tool used the most input tokens over MCP**

Claude Code (Opus 5.5, medium). The mark is each setup’s median of 12 runs, and each tick on its line is one run. Lower is better.

Dot-and-range chart. With Claude Code, every tool used the most input tokens over MCP. Input tokens per run, agent-browser, CLI + skill: its skill loads a 38 thousand-character guide. Input tokens per run, agent-browser, MCP: the most of all 18 setups. Time per run, Chrome DevTools, MCP: 10 of its 12 runs were faster than every run of the other two. The values follow in a table.

Axis: 0, 1M, 2M

Axis: 0, 50s, 100s, 150s

**Show the Claude Code values**

| Setup | Input tokens per run, median | lowest | highest | Time per run, median | lowest | highest |
| --- | --- | --- | --- | --- | --- | --- |
| agent-browser, CLI + skill | 652K | 472K | 1.09M | 104s | 82s | 142s |
| Playwright, CLI + skill | 332K | 257K | 400K | 90s | 74s | 128s |
| Chrome DevTools, CLI + skill | 341K | 283K | 508K | 97s | 79s | 120s |
| agent-browser, CLI, no skill | 304K | 244K | 395K | 96s | 73s | 126s |
| Playwright, CLI, no skill | 353K | 251K | 410K | 99s | 80s | 125s |
| Chrome DevTools, CLI, no skill | 348K | 288K | 401K | 98s | 81s | 116s |
| agent-browser, MCP | 1.59M | 1.29M | 1.99M | 106s | 87s | 170s |
| Playwright, MCP | 513K | 417K | 691K | 98s | 87s | 119s |
| Chrome DevTools, MCP | 517K | 455K | 592K | 72s | 62s | 112s |

Note: Input tokens include cache reads. Time runs from launching the harness to its final message. Each measure and each harness has its own scale. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

**With Codex, agent-browser used the most input tokens over every connection**

Codex (GPT 6 Astra, medium). The mark is each setup’s median of 12 runs, and each tick on its line is one run. Lower is better.

Dot-and-range chart. With Codex, agent-browser used the most input tokens over every connection. Input tokens per run, agent-browser, CLI + skill: its skill loads the same guide. Time per run, Chrome DevTools, CLI + skill: its 4 Windows runs took 140 to 168 seconds. The values follow in a table.

Axis: 0, 200K, 400K

Axis: 0, 100s, 200s

**Show the Codex values**

| Setup | Input tokens per run, median | lowest | highest | Time per run, median | lowest | highest |
| --- | --- | --- | --- | --- | --- | --- |
| agent-browser, CLI + skill | 426K | 278K | 505K | 91s | 70s | 127s |
| Playwright, CLI + skill | 318K | 211K | 399K | 102s | 70s | 151s |
| Chrome DevTools, CLI + skill | 197K | 173K | 257K | 75s | 68s | 168s |
| agent-browser, CLI, no skill | 289K | 228K | 335K | 78s | 59s | 107s |
| Playwright, CLI, no skill | 243K | 209K | 368K | 91s | 74s | 160s |
| Chrome DevTools, CLI, no skill | 233K | 207K | 283K | 86s | 69s | 277s |
| agent-browser, MCP | 414K | 333K | 473K | 72s | 59s | 86s |
| Playwright, MCP | 261K | 188K | 335K | 88s | 66s | 117s |
| Chrome DevTools, MCP | 238K | 166K | 365K | 81s | 60s | 112s |

Note: Pooled across the three OSes, run times overlapped in every Codex matchup, so no tool was clearly faster. Each measure and each harness has its own scale. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

## The job

**Two harnesses, three connections and three tools make 18 setups**

Grid of the 18 setups. Two harnesses, three connections and three tools make 18 setups. Rows are the harnesses, Claude Code (Opus 5.5, medium) and Codex (GPT 6 Astra, medium). Columns are the connections: CLI + skill, the tool’s CLI and its official skill; CLI, no skill, the same CLI without the skill; MCP, the tool’s MCP server, no skill. Each cell is a matchup of the three tools, and each tool in it is a setup: agent-browser by Vercel Labs, Playwright by Microsoft, Chrome DevTools by Google. Each setup ran 12 times, 4 on each of macOS, WSL, Windows.

**Every run did the same steps in a signed-in web app**

1. Pick the right option from a menu.
2. Open a report in a new tab and check it.
3. Read its hundreds of rows and save them.
4. Close the report’s tab.
5. Go back to the start page.

## Speed

**Only Chrome DevTools over MCP with Claude Code was clearly faster than the other two tools**

Every run’s time with Claude Code (Opus 5.5, medium) and Codex (GPT 6 Astra, medium), 12 runs per tool in each matchup. The other 5 matchups are in grey.

Strip chart of every run’s time, one dot per run, with a panel for each of the 6 matchups. Only Chrome DevTools over MCP with Claude Code was clearly faster than the other two tools. The time axis starts at 40 seconds, not at zero. Each run’s values are in the table of every run at the end of this post.

[Skip the chart of every run](https://maximilianfabian.dev/blog/browser-tools-for-agents/#runs-end)

### Claude Code, MCP

agent-browser 87 to 170 seconds, Playwright 87 to 119 seconds, Chrome DevTools 62 to 112 seconds. Chrome DevTools: 10 of its 12 runs were faster than every run of the other two tools in this matchup, left of the dashed line.

agent-browser

macOS, 105 seconds  
macOS, 107 seconds  
macOS, 102 seconds  
macOS, 97 seconds  
WSL, 131 seconds  
WSL, 96 seconds  
WSL, 87 seconds  
WSL, 102 seconds  
Windows, 133 seconds  
Windows, 144 seconds  
Windows, 170 seconds  
Windows, 168 seconds

Playwright

macOS, 108 seconds  
macOS, 112 seconds  
macOS, 91 seconds  
macOS, 119 seconds  
WSL, 87 seconds  
WSL, 93 seconds  
WSL, 116 seconds  
WSL, 93 seconds  
Windows, 103 seconds  
Windows, 91 seconds  
Windows, 93 seconds  
Windows, 105 seconds

Chrome DevTools

macOS, 64 seconds  
macOS, 85 seconds  
macOS, 66 seconds  
macOS, 70 seconds  
WSL, 62 seconds  
WSL, 84 seconds  
WSL, 69 seconds  
WSL, 107 seconds  
Windows, 79 seconds  
Windows, 112 seconds  
Windows, 69 seconds  
Windows, 73 seconds

### Claude Code, CLI + skill

agent-browser 82 to 142 seconds, Playwright 74 to 128 seconds, Chrome DevTools 79 to 120 seconds.

agent-browser

macOS, 117 seconds  
macOS, 97 seconds  
macOS, 142 seconds  
macOS, 132 seconds  
WSL, 96 seconds  
WSL, 82 seconds  
WSL, 84 seconds  
WSL, 96 seconds  
Windows, 129 seconds  
Windows, 110 seconds  
Windows, 93 seconds  
Windows, 120 seconds

Playwright

macOS, 128 seconds  
macOS, 78 seconds  
macOS, 91 seconds  
macOS, 79 seconds  
WSL, 85 seconds  
WSL, 102 seconds  
WSL, 77 seconds  
WSL, 74 seconds  
Windows, 100 seconds  
Windows, 89 seconds  
Windows, 94 seconds  
Windows, 100 seconds

Chrome DevTools

macOS, 79 seconds  
macOS, 79 seconds  
macOS, 79 seconds  
macOS, 112 seconds  
WSL, 119 seconds  
WSL, 94 seconds  
WSL, 79 seconds  
WSL, 85 seconds  
Windows, 116 seconds  
Windows, 120 seconds  
Windows, 100 seconds  
Windows, 118 seconds

### Claude Code, CLI, no skill

agent-browser 73 to 126 seconds, Playwright 80 to 125 seconds, Chrome DevTools 81 to 116 seconds.

agent-browser

macOS, 102 seconds  
macOS, 89 seconds  
macOS, 100 seconds  
macOS, 103 seconds  
WSL, 92 seconds  
WSL, 73 seconds  
WSL, 91 seconds  
WSL, 76 seconds  
Windows, 107 seconds  
Windows, 83 seconds  
Windows, 126 seconds  
Windows, 122 seconds

Playwright

macOS, 97 seconds  
macOS, 85 seconds  
macOS, 102 seconds  
macOS, 80 seconds  
WSL, 94 seconds  
WSL, 95 seconds  
WSL, 116 seconds  
WSL, 97 seconds  
Windows, 108 seconds  
Windows, 125 seconds  
Windows, 101 seconds  
Windows, 116 seconds

Chrome DevTools

macOS, 98 seconds  
macOS, 84 seconds  
macOS, 84 seconds  
macOS, 89 seconds  
WSL, 97 seconds  
WSL, 110 seconds  
WSL, 85 seconds  
WSL, 81 seconds  
Windows, 102 seconds  
Windows, 111 seconds  
Windows, 116 seconds  
Windows, 106 seconds

### Codex, CLI + skill

agent-browser 70 to 127 seconds, Playwright 70 to 151 seconds, Chrome DevTools 68 to 168 seconds.

agent-browser

macOS, 90 seconds  
macOS, 95 seconds  
macOS, 94 seconds  
macOS, 84 seconds  
WSL, 79 seconds  
WSL, 70 seconds  
WSL, 75 seconds  
WSL, 127 seconds  
Windows, 92 seconds  
Windows, 106 seconds  
Windows, 105 seconds  
Windows, 91 seconds

Playwright

macOS, 106 seconds  
macOS, 92 seconds  
macOS, 94 seconds  
macOS, 70 seconds  
WSL, 104 seconds  
WSL, 99 seconds  
WSL, 90 seconds  
WSL, 78 seconds  
Windows, 137 seconds  
Windows, 151 seconds  
Windows, 127 seconds  
Windows, 128 seconds

Chrome DevTools

macOS, 68 seconds  
macOS, 69 seconds  
macOS, 73 seconds  
macOS, 70 seconds  
WSL, 76 seconds  
WSL, 74 seconds  
WSL, 78 seconds  
WSL, 72 seconds  
Windows, 140 seconds  
Windows, 143 seconds  
Windows, 142 seconds  
Windows, 168 seconds

### Codex, CLI, no skill

agent-browser 59 to 107 seconds, Playwright 74 to 160 seconds, Chrome DevTools 69 to 277 seconds.

agent-browser

macOS, 91 seconds  
macOS, 68 seconds  
macOS, 77 seconds  
macOS, 76 seconds  
WSL, 59 seconds  
WSL, 78 seconds  
WSL, 59 seconds  
WSL, 69 seconds  
Windows, 90 seconds  
Windows, 87 seconds  
Windows, 107 seconds  
Windows, 95 seconds

Playwright

macOS, 84 seconds  
macOS, 74 seconds  
macOS, 97 seconds  
macOS, 106 seconds  
WSL, 82 seconds  
WSL, 81 seconds  
WSL, 79 seconds  
WSL, 82 seconds  
Windows, 136 seconds  
Windows, 125 seconds  
Windows, 160 seconds  
Windows, 153 seconds

Chrome DevTools

macOS, 74 seconds  
macOS, 87 seconds  
macOS, 85 seconds  
macOS, 71 seconds  
WSL, 107 seconds  
WSL, 69 seconds  
WSL, 80 seconds  
WSL, 74 seconds  
Windows, 145 seconds  
Windows, 138 seconds  
Windows, 169 seconds  
Windows, 277 seconds  
277s

### Codex, MCP

agent-browser 59 to 86 seconds, Playwright 66 to 117 seconds, Chrome DevTools 60 to 112 seconds.

agent-browser

macOS, 86 seconds  
macOS, 76 seconds  
macOS, 77 seconds  
macOS, 67 seconds  
WSL, 68 seconds  
WSL, 61 seconds  
WSL, 62 seconds  
WSL, 64 seconds  
Windows, 82 seconds  
Windows, 77 seconds  
Windows, 80 seconds  
Windows, 59 seconds

Playwright

macOS, 117 seconds  
macOS, 113 seconds  
macOS, 72 seconds  
macOS, 66 seconds  
WSL, 89 seconds  
WSL, 79 seconds  
WSL, 84 seconds  
WSL, 70 seconds  
Windows, 97 seconds  
Windows, 111 seconds  
Windows, 90 seconds  
Windows, 87 seconds

Chrome DevTools

macOS, 60 seconds  
macOS, 82 seconds  
macOS, 64 seconds  
macOS, 89 seconds  
WSL, 81 seconds  
WSL, 63 seconds  
WSL, 67 seconds  
WSL, 66 seconds  
Windows, 93 seconds  
Windows, 112 seconds  
Windows, 88 seconds  
Windows, 84 seconds

Note: All panels share one time axis, which starts at 40s. A run past 200s sits on its right edge and shows its time. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

**Most of each run was waiting on the model**

Median time per run in seconds, 12 runs per setup. Both harnesses use one scale.

Bar chart of each setup’s median time per run, with the part inside the browser tool. Most of each run was waiting on the model. Both harnesses use one scale. The values follow in a table.

| Setup | Median time per run | Part inside the browser tool |
| --- | --- | --- |
| Claude Code, agent-browser, CLI + skill | 104 seconds | 11 seconds |
| Claude Code, agent-browser, CLI, no skill | 96 seconds | 14 seconds |
| Claude Code, agent-browser, MCP | 106 seconds | 4 seconds |
| Claude Code, Playwright, CLI + skill | 90 seconds | 18 seconds |
| Claude Code, Playwright, CLI, no skill | 99 seconds | 16 seconds |
| Claude Code, Playwright, MCP | 98 seconds | 9 seconds |
| Claude Code, Chrome DevTools, CLI + skill | 97 seconds | 14 seconds |
| Claude Code, Chrome DevTools, CLI, no skill | 98 seconds | 17 seconds |
| Claude Code, Chrome DevTools, MCP | 72 seconds | 4 seconds |
| Codex, agent-browser, CLI + skill | 91 seconds | 3 seconds |
| Codex, agent-browser, CLI, no skill | 78 seconds | 3 seconds |
| Codex, agent-browser, MCP | 72 seconds | 3 seconds |
| Codex, Playwright, CLI + skill | 102 seconds | 12 seconds |
| Codex, Playwright, CLI, no skill | 91 seconds | 12 seconds |
| Codex, Playwright, MCP | 88 seconds | 13 seconds |
| Codex, Chrome DevTools, CLI + skill | 75 seconds | 9 seconds |
| Codex, Chrome DevTools, CLI, no skill | 86 seconds | 10 seconds |
| Codex, Chrome DevTools, MCP | 81 seconds | 3 seconds |

Note: The browser tool took 3 to 18s of each setup’s median run. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

Since every run passed its checks, I’d choose a tool by its token use and speed.

### Split by OS

The charts above combine the three operating systems. Split by OS, each setup has 4 runs. I compared each pair of tools within each harness, connection and OS. In 15 of those 54 comparisons, every run of one tool was faster than every run of the other, and 9 of the 15 were on Windows.

**With Claude Code, only Chrome DevTools over MCP on macOS was faster than both other tools**

Claude Code (Opus 5.5, medium). Each dot is one run, 4 per tool in each panel, and further left is faster. A panel is in colour when every run of one tool was faster than every run of both others, and a dashed line marks the fastest run of the other two. A note under a panel names any tool whose every run was faster than every run of another.

Strip chart of Claude Code’s run times, one dot per run, with a panel per connection and OS. With Claude Code, only Chrome DevTools over MCP on macOS was faster than both other tools. The time axis starts at 40 seconds, not at zero. Each run’s values are in the table of every run at the end of this post.

[Skip the chart of Claude Code’s runs on each OS](https://maximilianfabian.dev/blog/browser-tools-for-agents/#runs-end-claude-code)

#### Claude Code, CLI + skill, macOS

agent-browser 97 to 142 seconds, Playwright 78 to 128 seconds, Chrome DevTools 79 to 112 seconds.

agent-browser

agent-browser, 117 seconds  
agent-browser, 97 seconds  
agent-browser, 142 seconds  
agent-browser, 132 seconds

Playwright

Playwright, 128 seconds  
Playwright, 78 seconds  
Playwright, 91 seconds  
Playwright, 79 seconds

Chrome DevTools

Chrome DevTools, 79 seconds  
Chrome DevTools, 79 seconds  
Chrome DevTools, 79 seconds  
Chrome DevTools, 112 seconds

#### Claude Code, CLI + skill, WSL

agent-browser 82 to 96 seconds, Playwright 74 to 102 seconds, Chrome DevTools 79 to 119 seconds.

agent-browser

agent-browser, 96 seconds  
agent-browser, 82 seconds  
agent-browser, 84 seconds  
agent-browser, 96 seconds

Playwright

Playwright, 85 seconds  
Playwright, 102 seconds  
Playwright, 77 seconds  
Playwright, 74 seconds

Chrome DevTools

Chrome DevTools, 119 seconds  
Chrome DevTools, 94 seconds  
Chrome DevTools, 79 seconds  
Chrome DevTools, 85 seconds

#### Claude Code, CLI + skill, Windows

agent-browser 93 to 129 seconds, Playwright 89 to 100 seconds, Chrome DevTools 100 to 120 seconds. Playwright was faster than Chrome DevTools.

agent-browser

agent-browser, 129 seconds  
agent-browser, 110 seconds  
agent-browser, 93 seconds  
agent-browser, 120 seconds

Playwright

Playwright, 100 seconds  
Playwright, 89 seconds  
Playwright, 94 seconds  
Playwright, 100 seconds

Chrome DevTools

Chrome DevTools, 116 seconds  
Chrome DevTools, 120 seconds  
Chrome DevTools, 100 seconds  
Chrome DevTools, 118 seconds

#### Claude Code, CLI, no skill, macOS

agent-browser 89 to 103 seconds, Playwright 80 to 102 seconds, Chrome DevTools 84 to 98 seconds.

agent-browser

agent-browser, 102 seconds  
agent-browser, 89 seconds  
agent-browser, 100 seconds  
agent-browser, 103 seconds

Playwright

Playwright, 97 seconds  
Playwright, 85 seconds  
Playwright, 102 seconds  
Playwright, 80 seconds

Chrome DevTools

Chrome DevTools, 98 seconds  
Chrome DevTools, 84 seconds  
Chrome DevTools, 84 seconds  
Chrome DevTools, 89 seconds

#### Claude Code, CLI, no skill, WSL

agent-browser 73 to 92 seconds, Playwright 94 to 116 seconds, Chrome DevTools 81 to 110 seconds. agent-browser was faster than Playwright.

agent-browser

agent-browser, 92 seconds  
agent-browser, 73 seconds  
agent-browser, 91 seconds  
agent-browser, 76 seconds

Playwright

Playwright, 94 seconds  
Playwright, 95 seconds  
Playwright, 116 seconds  
Playwright, 97 seconds

Chrome DevTools

Chrome DevTools, 97 seconds  
Chrome DevTools, 110 seconds  
Chrome DevTools, 85 seconds  
Chrome DevTools, 81 seconds

#### Claude Code, CLI, no skill, Windows

agent-browser 83 to 126 seconds, Playwright 101 to 125 seconds, Chrome DevTools 102 to 116 seconds.

agent-browser

agent-browser, 107 seconds  
agent-browser, 83 seconds  
agent-browser, 126 seconds  
agent-browser, 122 seconds

Playwright

Playwright, 108 seconds  
Playwright, 125 seconds  
Playwright, 101 seconds  
Playwright, 116 seconds

Chrome DevTools

Chrome DevTools, 102 seconds  
Chrome DevTools, 111 seconds  
Chrome DevTools, 116 seconds  
Chrome DevTools, 106 seconds

#### Claude Code, MCP, macOS

agent-browser 97 to 107 seconds, Playwright 91 to 119 seconds, Chrome DevTools 64 to 85 seconds. Chrome DevTools was faster than agent-browser and Playwright.

agent-browser

agent-browser, 105 seconds  
agent-browser, 107 seconds  
agent-browser, 102 seconds  
agent-browser, 97 seconds

Playwright

Playwright, 108 seconds  
Playwright, 112 seconds  
Playwright, 91 seconds  
Playwright, 119 seconds

Chrome DevTools

Chrome DevTools, 64 seconds  
Chrome DevTools, 85 seconds  
Chrome DevTools, 66 seconds  
Chrome DevTools, 70 seconds

#### Claude Code, MCP, WSL

agent-browser 87 to 131 seconds, Playwright 87 to 116 seconds, Chrome DevTools 62 to 107 seconds.

agent-browser

agent-browser, 131 seconds  
agent-browser, 96 seconds  
agent-browser, 87 seconds  
agent-browser, 102 seconds

Playwright

Playwright, 87 seconds  
Playwright, 93 seconds  
Playwright, 116 seconds  
Playwright, 93 seconds

Chrome DevTools

Chrome DevTools, 62 seconds  
Chrome DevTools, 84 seconds  
Chrome DevTools, 69 seconds  
Chrome DevTools, 107 seconds

#### Claude Code, MCP, Windows

agent-browser 133 to 170 seconds, Playwright 91 to 105 seconds, Chrome DevTools 69 to 112 seconds. Playwright and Chrome DevTools were faster than agent-browser.

agent-browser

agent-browser, 133 seconds  
agent-browser, 144 seconds  
agent-browser, 170 seconds  
agent-browser, 168 seconds

Playwright

Playwright, 103 seconds  
Playwright, 91 seconds  
Playwright, 93 seconds  
Playwright, 105 seconds

Chrome DevTools

Chrome DevTools, 79 seconds  
Chrome DevTools, 112 seconds  
Chrome DevTools, 69 seconds  
Chrome DevTools, 73 seconds

Note: All panels share one time axis, which starts at 40s. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

**With Codex on Windows, agent-browser was faster than both other tools over every connection**

Codex (GPT 6 Astra, medium). Each dot is one run, 4 per tool in each panel, and further left is faster. A panel is in colour when every run of one tool was faster than every run of both others, and a dashed line marks the fastest run of the other two. A note under a panel names any tool whose every run was faster than every run of another.

Strip chart of Codex’s run times, one dot per run, with a panel per connection and OS. With Codex on Windows, agent-browser was faster than both other tools over every connection. The time axis starts at 40 seconds, not at zero. Each run’s values are in the table of every run at the end of this post.

[Skip the chart of Codex’s runs on each OS](https://maximilianfabian.dev/blog/browser-tools-for-agents/#runs-end-codex)

#### Codex, CLI + skill, macOS

agent-browser 84 to 95 seconds, Playwright 70 to 106 seconds, Chrome DevTools 68 to 73 seconds. Chrome DevTools was faster than agent-browser.

agent-browser

agent-browser, 90 seconds  
agent-browser, 95 seconds  
agent-browser, 94 seconds  
agent-browser, 84 seconds

Playwright

Playwright, 106 seconds  
Playwright, 92 seconds  
Playwright, 94 seconds  
Playwright, 70 seconds

Chrome DevTools

Chrome DevTools, 68 seconds  
Chrome DevTools, 69 seconds  
Chrome DevTools, 73 seconds  
Chrome DevTools, 70 seconds

#### Codex, CLI + skill, WSL

agent-browser 70 to 127 seconds, Playwright 78 to 104 seconds, Chrome DevTools 72 to 78 seconds.

agent-browser

agent-browser, 79 seconds  
agent-browser, 70 seconds  
agent-browser, 75 seconds  
agent-browser, 127 seconds

Playwright

Playwright, 104 seconds  
Playwright, 99 seconds  
Playwright, 90 seconds  
Playwright, 78 seconds

Chrome DevTools

Chrome DevTools, 76 seconds  
Chrome DevTools, 74 seconds  
Chrome DevTools, 78 seconds  
Chrome DevTools, 72 seconds

#### Codex, CLI + skill, Windows

agent-browser 91 to 106 seconds, Playwright 127 to 151 seconds, Chrome DevTools 140 to 168 seconds. agent-browser was faster than Playwright and Chrome DevTools.

agent-browser

agent-browser, 92 seconds  
agent-browser, 106 seconds  
agent-browser, 105 seconds  
agent-browser, 91 seconds

Playwright

Playwright, 137 seconds  
Playwright, 151 seconds  
Playwright, 127 seconds  
Playwright, 128 seconds

Chrome DevTools

Chrome DevTools, 140 seconds  
Chrome DevTools, 143 seconds  
Chrome DevTools, 142 seconds  
Chrome DevTools, 168 seconds

#### Codex, CLI, no skill, macOS

agent-browser 68 to 91 seconds, Playwright 74 to 106 seconds, Chrome DevTools 71 to 87 seconds.

agent-browser

agent-browser, 91 seconds  
agent-browser, 68 seconds  
agent-browser, 77 seconds  
agent-browser, 76 seconds

Playwright

Playwright, 84 seconds  
Playwright, 74 seconds  
Playwright, 97 seconds  
Playwright, 106 seconds

Chrome DevTools

Chrome DevTools, 74 seconds  
Chrome DevTools, 87 seconds  
Chrome DevTools, 85 seconds  
Chrome DevTools, 71 seconds

#### Codex, CLI, no skill, WSL

agent-browser 59 to 78 seconds, Playwright 79 to 82 seconds, Chrome DevTools 69 to 107 seconds. agent-browser was faster than Playwright.

agent-browser

agent-browser, 59 seconds  
agent-browser, 78 seconds  
agent-browser, 59 seconds  
agent-browser, 69 seconds

Playwright

Playwright, 82 seconds  
Playwright, 81 seconds  
Playwright, 79 seconds  
Playwright, 82 seconds

Chrome DevTools

Chrome DevTools, 107 seconds  
Chrome DevTools, 69 seconds  
Chrome DevTools, 80 seconds  
Chrome DevTools, 74 seconds

#### Codex, CLI, no skill, Windows

agent-browser 87 to 107 seconds, Playwright 125 to 160 seconds, Chrome DevTools 138 to 277 seconds. agent-browser was faster than Playwright and Chrome DevTools.

agent-browser

agent-browser, 90 seconds  
agent-browser, 87 seconds  
agent-browser, 107 seconds  
agent-browser, 95 seconds

Playwright

Playwright, 136 seconds  
Playwright, 125 seconds  
Playwright, 160 seconds  
Playwright, 153 seconds

Chrome DevTools

Chrome DevTools, 145 seconds  
Chrome DevTools, 138 seconds  
Chrome DevTools, 169 seconds  
Chrome DevTools, 277 seconds  
277s

#### Codex, MCP, macOS

agent-browser 67 to 86 seconds, Playwright 66 to 117 seconds, Chrome DevTools 60 to 89 seconds.

agent-browser

agent-browser, 86 seconds  
agent-browser, 76 seconds  
agent-browser, 77 seconds  
agent-browser, 67 seconds

Playwright

Playwright, 117 seconds  
Playwright, 113 seconds  
Playwright, 72 seconds  
Playwright, 66 seconds

Chrome DevTools

Chrome DevTools, 60 seconds  
Chrome DevTools, 82 seconds  
Chrome DevTools, 64 seconds  
Chrome DevTools, 89 seconds

#### Codex, MCP, WSL

agent-browser 61 to 68 seconds, Playwright 70 to 89 seconds, Chrome DevTools 63 to 81 seconds. agent-browser was faster than Playwright.

agent-browser

agent-browser, 68 seconds  
agent-browser, 61 seconds  
agent-browser, 62 seconds  
agent-browser, 64 seconds

Playwright

Playwright, 89 seconds  
Playwright, 79 seconds  
Playwright, 84 seconds  
Playwright, 70 seconds

Chrome DevTools

Chrome DevTools, 81 seconds  
Chrome DevTools, 63 seconds  
Chrome DevTools, 67 seconds  
Chrome DevTools, 66 seconds

#### Codex, MCP, Windows

agent-browser 59 to 82 seconds, Playwright 87 to 111 seconds, Chrome DevTools 84 to 112 seconds. agent-browser was faster than Playwright and Chrome DevTools.

agent-browser

agent-browser, 82 seconds  
agent-browser, 77 seconds  
agent-browser, 80 seconds  
agent-browser, 59 seconds

Playwright

Playwright, 97 seconds  
Playwright, 111 seconds  
Playwright, 90 seconds  
Playwright, 87 seconds

Chrome DevTools

Chrome DevTools, 93 seconds  
Chrome DevTools, 112 seconds  
Chrome DevTools, 88 seconds  
Chrome DevTools, 84 seconds

Note: All panels share one time axis, which starts at 40s. A run past 200s sits on its right edge and shows its time. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

With Codex, agent-browser still used the most input tokens over its CLI with the skill on every OS. These results come from 4 runs per setup on each OS, on one Mac and one PC, so a rerun or another machine might not match them.

## Token use over MCP

agent-browser over MCP with Claude Code was the most expensive of all 18 setups. It made about as many model calls as over its CLI with the skill, but each call used more input tokens.

**Each tool’s model calls were largest over MCP with Claude Code**

Median input tokens per model call, 12 runs per setup. Both harnesses use one scale.

Bar chart of each setup’s median input tokens per model call. Each tool’s model calls were largest over MCP with Claude Code. Both harnesses use one scale. The values follow in a table.

| Setup | Median input tokens per model call |
| --- | --- |
| Claude Code, agent-browser, CLI + skill | 30.1 thousand |
| Claude Code, agent-browser, CLI, no skill | 15.9 thousand |
| Claude Code, agent-browser, MCP | 72.7 thousand |
| Claude Code, Playwright, CLI + skill | 20.7 thousand |
| Claude Code, Playwright, CLI, no skill | 17.6 thousand |
| Claude Code, Playwright, MCP | 24.5 thousand |
| Claude Code, Chrome DevTools, CLI + skill | 19.5 thousand |
| Claude Code, Chrome DevTools, CLI, no skill | 18.6 thousand |
| Claude Code, Chrome DevTools, MCP | 28.8 thousand |
| Codex, agent-browser, CLI + skill | 30.4 thousand |
| Codex, agent-browser, CLI, no skill | 19.9 thousand |
| Codex, agent-browser, MCP | 31.9 thousand |
| Codex, Playwright, CLI + skill | 20.6 thousand |
| Codex, Playwright, CLI, no skill | 17.2 thousand |
| Codex, Playwright, MCP | 18.9 thousand |
| Codex, Chrome DevTools, CLI + skill | 18.3 thousand |
| Codex, Chrome DevTools, CLI, no skill | 17.9 thousand |
| Codex, Chrome DevTools, MCP | 19.6 thousand |

Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

With Codex, no connection was consistently cheaper.

## agent-browser’s skill

agent-browser’s skill is a stub: it tells the agent to run `agent-browser skills get core`, which prints a long guide that stays in context for the rest of the run.

**agent-browser’s skill puts the most in the agent’s context**

Each square is 1,000 characters.

Unit chart, one square per 1,000 characters. agent-browser’s skill is the smallest, at 3.5 thousand characters, but it loads a 38 thousand-character guide, so it puts the most in the agent’s context. Playwright’s skill is 15 thousand and Chrome DevTools’ is 12 thousand.

| Tool | Characters in its skill | Characters in the guide its skill loads |
| --- | --- | --- |
| agent-browser | 3,455 | 37,980 |
| Playwright | 15,422 | None |
| Chrome DevTools | 12,031 | None |

Source: [the SKILL.md in each tool’s official skill](https://maximilianfabian.dev/blog/browser-tools-for-agents/#how-i-ran-the-study) at the version I tested, and the guide that agent-browser’s skill loads, 19 to 22 October 2026.

Over its CLI, dropping the skill cut agent-browser’s input tokens by 53% with Claude Code and 32% with Codex, while the other two tools’ CLI runs showed no consistent gap. In 67 of the 72 CLI runs without a skill, the agent read the tool’s `--help` instead. For a job like this one, I’d try agent-browser without its skill.

## A script instead of an agent

Between agent runs, a Python script did the same job over CDP, using no model calls and no tokens at run time.

**The script’s slowest run was faster than the fastest agent run**

Time of every run in seconds: 258 runs of the script, and 108 agent runs with each harness.

Strip chart of every run’s time on one axis from zero. The script’s slowest run was faster than the fastest agent run. The values follow in a table.

Axis: 0, 100s, 200s

| What ran | Runs | Fastest | Median | Slowest |
| --- | --- | --- | --- | --- |
| Python script over CDP | 258 | 1.4 seconds | 1.5 seconds | 10.3 seconds |
| Claude Code (Opus 5.5, medium) | 108 | 62 seconds | 97 seconds | 170 seconds |
| Codex (GPT 6 Astra, medium) | 108 | 59 seconds | 84 seconds | 277 seconds |

Note: The script’s times don’t include writing it. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

An earlier Claude Code session did this job with agent-browser and left notes, and the session I built the study with turned those notes into the script. Then I broke the script on purpose, and an agent running it finished the job itself and proposed three fixes, which went into the script.

**For a job you repeat, I’d let an agent write a script and call the agent back when the script breaks**

1. An agent does the job once and keeps notes.
2. An agent turns the notes into a script.
3. The script runs the job on its own.
4. When the script breaks, an agent finishes the job and fixes the script.

## Windows and WSL

**Windows runs spent the most time inside the browser tool**

Median time per run in seconds, 36 runs per harness on each OS. Both harnesses use one scale.

Column chart of median time per run on each operating system, with the part spent inside the browser tool. Runs were slowest on Windows. 36 runs each. Both harnesses use one scale. The values follow in a table.

| Harness and OS | Median time per run | Part inside the browser tool |
| --- | --- | --- |
| Claude Code, macOS | 97 seconds | 13 seconds |
| Claude Code, WSL | 92 seconds | 10 seconds |
| Claude Code, Windows | 107 seconds | 15 seconds |
| Codex, macOS | 79 seconds | 8 seconds |
| Codex, WSL | 77 seconds | 7 seconds |
| Codex, Windows | 109 seconds | 24 seconds |

Note: WSL and Windows ran on the same PC. Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

**Each tool’s CLI commands took longest on Windows**

Median seconds per CLI command, over both harnesses’ CLI runs on each OS.

Bar chart of the median time per CLI command for each tool on each operating system. Each tool’s CLI commands took longest on Windows. The values follow in a table.

| Tool and OS | Median time per CLI command |
| --- | --- |
| agent-browser, macOS | 0.2 seconds |
| agent-browser, WSL | 0.1 seconds |
| agent-browser, Windows | 0.5 seconds |
| Playwright, macOS | 0.7 seconds |
| Playwright, WSL | 0.7 seconds |
| Playwright, Windows | 1.4 seconds |
| Chrome DevTools, macOS | 0.5 seconds |
| Chrome DevTools, WSL | 0.6 seconds |
| Chrome DevTools, Windows | 1.9 seconds |

Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

On a Windows PC, I’d run agents in WSL.

## Limits

I measured one job in one web app, at medium effort, with 12 runs per setup, so another job could order the tools differently. I report each harness on its own because they ran different models at different prices.

The Windows and WSL result comes from one PC and the versions listed below. Every number pools the three OSes unless I split it.

## How I ran the study

**Show the method**

I used these versions:

- Claude Code 2.1.303 with `claude-opus-5-5`, medium effort
- Codex 0.163.0-alpha.11 with `gpt-6-astra`, medium effort
- agent-browser 0.39.1 (Vercel Labs), the CLI and its built-in MCP server
- Playwright CLI 0.1.22 and Playwright MCP 0.0.83 (Microsoft)
- Chrome DevTools for agents 1.11.0 (Google), its `chrome-devtools` CLI and `chrome-devtools-mcp` server
- Google Chrome 155, stable release, headless, with one study profile per operating system

The CLI + skill setups got each tool’s official skill folder, unchanged, from its GitHub repository at the version above: [agent-browser](https://github.com/vercel-labs/agent-browser/tree/v0.39.1/skills/agent-browser), [Playwright CLI](https://github.com/microsoft/playwright-cli/tree/v0.1.22/skills/playwright-cli) and [Chrome DevTools](https://github.com/ChromeDevTools/chrome-devtools-mcp/tree/chrome-devtools-mcp-v1.11.0/skills/chrome-devtools-cli). Each tool’s documentation tells you to install that folder.

A run passed only if it saved rows that matched a reference exactly, left the app in the right state, used only its assigned browser tool, never asked me for input and finished within 30 minutes.

I ran the study from 19 to 22 October 2026. Every run was a fresh harness session with empty settings, no other skills or MCP servers and no project instruction files. Each got the same instructions, started with the app open and signed in, and read the report’s table with the same JavaScript. Each number is a setup’s median over its 12 runs unless I say otherwise. I took each measure’s median on its own, so a median run can mix values from different runs.

Token counts come from each harness’s own logs. Cache reads made up 93.9% of input tokens with Claude Code and 90.5% with Codex. The tables also show output tokens, which include reasoning.

Claude Code reports what each run would cost through the API, and I priced Codex’s tokens at list prices. Neither is what you’d pay on a subscription plan. Within one harness, cost roughly follows tokens, so I chart tokens and give costs in the text and the tables.

The web app’s name and the report’s contents stay private, so this post reports only derived measurements.

## Every setup’s median run

**Claude Code (Opus 5.5, medium)**

12 runs per setup, 4 on each OS.

**Every setup’s median run with Claude Code (Opus 5.5, medium)**

| Connection | Time | Input tokens | Cost | In browser tool | Output tokens | Model calls | Browser commands |
| --- | --- | --- | --- | --- | --- | --- | --- |
| agent-browser |   |   |   |   |   |   |   |
| CLI + skill | 104s | 652K | $0.48 | 11s | 6.8K | 22 | 40 |
| CLI, no skill | 96s | 304K | $0.28 | 14s | 5.9K | 20 | 38 |
| MCP | 106s | 1.59M | $1.02 | 4s | 8.7K | 21 | 27 |
| Playwright |   |   |   |   |   |   |   |
| CLI + skill | 90s | 332K | $0.32 | 18s | 5.8K | 16 | 22 |
| CLI, no skill | 99s | 353K | $0.33 | 16s | 6.6K | 20 | 25 |
| MCP | 98s | 513K | $0.41 | 9s | 7.3K | 21 | 15 |
| Chrome DevTools |   |   |   |   |   |   |   |
| CLI + skill | 97s | 341K | $0.32 | 14s | 6.1K | 17 | 22 |
| CLI, no skill | 98s | 348K | $0.32 | 17s | 5.9K | 19 | 24 |
| MCP | 72s | 517K | $0.40 | 4s | 5.7K | 18 | 15 |

Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

**Codex (GPT 6 Astra, medium)**

12 runs per setup, 4 on each OS.

**Every setup’s median run with Codex (GPT 6 Astra, medium)**

| Connection | Time | Input tokens | Cost | In browser tool | Output tokens | Model calls | Browser commands |
| --- | --- | --- | --- | --- | --- | --- | --- |
| agent-browser |   |   |   |   |   |   |   |
| CLI + skill | 91s | 426K | $0.91 | 3s | 2.2K | 14 | 27 |
| CLI, no skill | 78s | 289K | $0.68 | 3s | 1.9K | 14 | 23 |
| MCP | 72s | 414K | $0.89 | 3s | 1.8K | 13 | 21 |
| Playwright |   |   |   |   |   |   |   |
| CLI + skill | 102s | 318K | $0.80 | 12s | 2.2K | 16 | 16 |
| CLI, no skill | 91s | 243K | $0.66 | 12s | 2.2K | 14 | 18 |
| MCP | 88s | 261K | $0.62 | 13s | 2.1K | 14 | 18 |
| Chrome DevTools |   |   |   |   |   |   |   |
| CLI + skill | 75s | 197K | $0.53 | 9s | 1.9K | 11 | 17 |
| CLI, no skill | 86s | 233K | $0.58 | 10s | 1.9K | 13 | 18 |
| MCP | 81s | 238K | $0.60 | 3s | 2.0K | 12 | 15 |

Source: [my study runs](https://maximilianfabian.dev/blog/browser-tools-for-agents/#every-run), 19 to 22 October 2026.

## Every run

One row per run, with time in seconds and cost in US dollars.

**Every run as CSV**

```csv
Harness,Tool,Connection,OS,Run,Time (s),Input tokens,Cached input,Output tokens,Cost (USD),Model calls,Browser commands
Claude Code,agent-browser,CLI + skill,macOS,C1,116.8,623265,584110,9156,0.4957,23,55
Claude Code,agent-browser,CLI + skill,macOS,C2,97.4,569199,534281,5713,0.4235,19,39
Claude Code,agent-browser,CLI + skill,macOS,C3,142.4,682163,646584,9044,0.5158,23,39
Claude Code,agent-browser,CLI + skill,macOS,C4,132.3,1093881,1048684,10628,0.676,29,52
Claude Code,agent-browser,CLI + skill,WSL,C1,95.7,471985,437935,6209,0.4054,17,36
Claude Code,agent-browser,CLI + skill,WSL,C2,82.2,543769,509800,6843,0.4364,18,41
Claude Code,agent-browser,CLI + skill,WSL,C3,83.9,486134,452068,6763,0.4238,17,24
Claude Code,agent-browser,CLI + skill,WSL,C4,95.8,734957,691765,6075,0.5036,20,28
Claude Code,agent-browser,CLI + skill,Windows,C1,129.2,681670,646381,6191,0.4579,22,43
Claude Code,agent-browser,CLI + skill,Windows,C2,109.9,706145,668782,8985,0.5375,22,52
Claude Code,agent-browser,CLI + skill,Windows,C3,92.7,594483,562471,6643,0.4338,21,46
Claude Code,agent-browser,CLI + skill,Windows,C4,119.6,962950,917408,9827,0.645,25,35
Claude Code,agent-browser,"CLI, no skill",macOS,C1,101.6,341587,321506,7700,0.3465,20,42
Claude Code,agent-browser,"CLI, no skill",macOS,C2,88.9,304730,288599,5877,0.2837,19,44
Claude Code,agent-browser,"CLI, no skill",macOS,C3,99.9,346352,329634,7201,0.3213,22,41
Claude Code,agent-browser,"CLI, no skill",macOS,C4,103,340204,323655,5568,0.2866,21,38
Claude Code,agent-browser,"CLI, no skill",WSL,C1,91.5,269273,254068,5683,0.2683,18,34
Claude Code,agent-browser,"CLI, no skill",WSL,C2,73,244496,229518,5356,0.2557,16,35
Claude Code,agent-browser,"CLI, no skill",WSL,C3,91.3,360465,343898,7731,0.334,23,33
Claude Code,agent-browser,"CLI, no skill",WSL,C4,76,294146,276951,6077,0.2863,18,34
Claude Code,agent-browser,"CLI, no skill",Windows,C1,106.8,395009,377081,7175,0.3369,23,40
Claude Code,agent-browser,"CLI, no skill",Windows,C2,83,304086,290226,5861,0.2816,19,40
Claude Code,agent-browser,"CLI, no skill",Windows,C3,126.1,301055,287879,5877,0.2781,20,37
Claude Code,agent-browser,"CLI, no skill",Windows,C4,121.8,270652,256583,5848,0.267,19,37
Claude Code,agent-browser,MCP,macOS,C1,105.4,1619531,1495600,8756,1.0939,21,27
Claude Code,agent-browser,MCP,macOS,C2,107.2,1403413,1277608,8323,1.051,19,27
Claude Code,agent-browser,MCP,macOS,C3,101.6,1545504,1438812,9155,1.0043,21,28
Claude Code,agent-browser,MCP,macOS,C4,97,1603418,1489693,8623,1.039,20,28
Claude Code,agent-browser,MCP,WSL,C1,130.9,1617843,1514370,8517,0.9905,23,21
Claude Code,agent-browser,MCP,WSL,C2,95.9,1498581,1394831,7947,0.9566,21,26
Claude Code,agent-browser,MCP,WSL,C3,86.7,1289803,1208208,8102,0.8116,20,20
Claude Code,agent-browser,MCP,WSL,C4,101.5,1577932,1492459,9168,0.9092,23,25
Claude Code,agent-browser,MCP,Windows,C1,133.1,1426308,1357518,9130,0.9788,20,30
Claude Code,agent-browser,MCP,Windows,C2,168.2,1904787,1792941,8486,1.0875,24,25
Claude Code,agent-browser,MCP,Windows,C3,143.8,1864586,1798407,8950,1.0503,26,26
Claude Code,agent-browser,MCP,Windows,C4,169.6,1992230,1881767,10233,1.1333,27,27
Claude Code,Playwright,CLI + skill,macOS,C1,127.5,399690,375665,6560,0.3542,18,28
Claude Code,Playwright,CLI + skill,macOS,C2,77.5,304364,278452,5187,0.289,15,23
Claude Code,Playwright,CLI + skill,macOS,C3,90.9,316991,297255,5279,0.2915,16,22
Claude Code,Playwright,CLI + skill,macOS,C4,78.8,328600,307213,6040,0.3169,16,18
Claude Code,Playwright,CLI + skill,WSL,C1,84.5,363244,339728,6221,0.3333,17,23
Claude Code,Playwright,CLI + skill,WSL,C2,101.6,381186,356266,5429,0.3322,16,20
Claude Code,Playwright,CLI + skill,WSL,C3,76.6,314550,294746,5618,0.2981,16,20
Claude Code,Playwright,CLI + skill,WSL,C4,74,256717,236700,4776,0.2707,13,20
Claude Code,Playwright,CLI + skill,Windows,C1,99.5,393098,365680,6947,0.3731,16,22
Claude Code,Playwright,CLI + skill,Windows,C2,89.3,282898,261932,4714,0.2755,14,20
Claude Code,Playwright,CLI + skill,Windows,C3,94.3,335328,313174,6791,0.3376,16,22
Claude Code,Playwright,CLI + skill,Windows,C4,100,379986,357800,7002,0.3509,18,22
Claude Code,Playwright,"CLI, no skill",macOS,C1,97.2,410283,386802,6966,0.3618,21,30
Claude Code,Playwright,"CLI, no skill",macOS,C2,85.4,251297,231793,6492,0.3015,15,18
Claude Code,Playwright,"CLI, no skill",macOS,C3,102.1,374628,353625,7407,0.3516,21,25
Claude Code,Playwright,"CLI, no skill",macOS,C4,79.9,292361,273627,5868,0.2935,17,21
Claude Code,Playwright,"CLI, no skill",WSL,C1,93.7,319087,299570,6488,0.315,18,22
Claude Code,Playwright,"CLI, no skill",WSL,C2,95.2,325868,308670,7380,0.3318,19,22
Claude Code,Playwright,"CLI, no skill",WSL,C3,116.4,365122,344874,6989,0.3378,20,24
Claude Code,Playwright,"CLI, no skill",WSL,C4,97.1,341101,322237,5763,0.3018,20,25
Claude Code,Playwright,"CLI, no skill",Windows,C1,107.6,387138,364969,7251,0.3528,21,27
Claude Code,Playwright,"CLI, no skill",Windows,C2,125.1,316374,300451,5270,0.2823,19,25
Claude Code,Playwright,"CLI, no skill",Windows,C3,101.2,373167,352375,6693,0.3367,20,21
Claude Code,Playwright,"CLI, no skill",Windows,C4,115.9,384501,364510,6187,0.3206,22,26
Claude Code,Playwright,MCP,macOS,C1,108.4,521899,487825,7236,0.4126,21,17
Claude Code,Playwright,MCP,macOS,C2,112.4,506912,472458,7584,0.4184,20,17
Claude Code,Playwright,MCP,macOS,C3,90.9,528905,495588,7384,0.4133,22,13
Claude Code,Playwright,MCP,macOS,C4,118.5,655019,621922,7202,0.4339,27,13
Claude Code,Playwright,MCP,WSL,C1,86.7,518440,499450,7166,0.4067,21,15
Claude Code,Playwright,MCP,WSL,C2,93.1,417396,386804,6597,0.3622,19,17
Claude Code,Playwright,MCP,WSL,C3,115.5,690872,665007,9646,0.5152,27,19
Claude Code,Playwright,MCP,WSL,C4,93.3,449468,416054,7157,0.3934,18,15
Claude Code,Playwright,MCP,Windows,C1,102.9,580763,546830,7959,0.4382,23,15
Claude Code,Playwright,MCP,Windows,C2,91.3,463217,430293,7551,0.4017,19,16
Claude Code,Playwright,MCP,Windows,C3,93.3,455884,423204,7787,0.4037,19,15
Claude Code,Playwright,MCP,Windows,C4,104.6,481955,449264,6662,0.3865,20,15
Claude Code,Chrome DevTools,CLI + skill,macOS,C1,78.9,332835,307057,6041,0.3111,17,21
Claude Code,Chrome DevTools,CLI + skill,macOS,C2,79.1,326953,306648,5782,0.3063,17,23
Claude Code,Chrome DevTools,CLI + skill,macOS,C3,78.9,348502,326433,6102,0.3254,17,21
Claude Code,Chrome DevTools,CLI + skill,macOS,C4,111.8,291502,271629,5977,0.301,15,18
Claude Code,Chrome DevTools,CLI + skill,WSL,C1,118.9,508343,482961,9068,0.4326,23,30
Claude Code,Chrome DevTools,CLI + skill,WSL,C2,94.2,312100,292979,4667,0.2753,16,19
Claude Code,Chrome DevTools,CLI + skill,WSL,C3,78.7,283404,264443,5767,0.2908,15,20
Claude Code,Chrome DevTools,CLI + skill,WSL,C4,85,378445,357446,6385,0.3319,19,22
Claude Code,Chrome DevTools,CLI + skill,Windows,C1,116.2,355723,335232,6265,0.3232,18,22
Claude Code,Chrome DevTools,CLI + skill,Windows,C2,119.8,367891,348403,5998,0.3154,19,21
Claude Code,Chrome DevTools,CLI + skill,Windows,C3,100.3,328594,308513,6534,0.3212,17,22
Claude Code,Chrome DevTools,CLI + skill,Windows,C4,117.8,498954,457626,6890,0.4643,18,25
Claude Code,Chrome DevTools,"CLI, no skill",macOS,C1,97.7,345847,326196,5925,0.3098,19,24
Claude Code,Chrome DevTools,"CLI, no skill",macOS,C2,84.3,387549,366560,6166,0.338,19,20
Claude Code,Chrome DevTools,"CLI, no skill",macOS,C3,83.9,304904,286439,6221,0.3018,18,24
Claude Code,Chrome DevTools,"CLI, no skill",macOS,C4,89.2,325593,304468,5299,0.3003,17,23
Claude Code,Chrome DevTools,"CLI, no skill",WSL,C1,97.3,342741,323253,5392,0.2933,19,22
Claude Code,Chrome DevTools,"CLI, no skill",WSL,C2,110.1,401020,380686,7951,0.3646,22,24
Claude Code,Chrome DevTools,"CLI, no skill",WSL,C3,84.9,373668,350745,5923,0.331,18,24
Claude Code,Chrome DevTools,"CLI, no skill",WSL,C4,80.9,389920,367817,5633,0.3245,19,24
Claude Code,Chrome DevTools,"CLI, no skill",Windows,C1,102.3,346841,326638,5875,0.3122,18,24
Claude Code,Chrome DevTools,"CLI, no skill",Windows,C2,110.9,288332,271164,5048,0.2694,17,19
Claude Code,Chrome DevTools,"CLI, no skill",Windows,C3,115.8,357968,336528,6806,0.3346,19,23
Claude Code,Chrome DevTools,"CLI, no skill",Windows,C4,105.6,349957,329889,6637,0.3274,19,21
Claude Code,Chrome DevTools,MCP,macOS,C1,64.2,518779,482736,5147,0.3797,18,13
Claude Code,Chrome DevTools,MCP,macOS,C2,84.6,514662,476908,5675,0.3976,18,14
Claude Code,Chrome DevTools,MCP,macOS,C3,66.3,467466,429759,5553,0.3855,16,15
Claude Code,Chrome DevTools,MCP,macOS,C4,69.8,454948,418393,5455,0.3755,16,14
Claude Code,Chrome DevTools,MCP,WSL,C1,61.9,461992,425869,5078,0.3673,16,14
Claude Code,Chrome DevTools,MCP,WSL,C2,83.8,590591,557451,6481,0.4068,22,18
Claude Code,Chrome DevTools,MCP,WSL,C3,69.2,535587,495244,5490,0.4105,18,15
Claude Code,Chrome DevTools,MCP,WSL,C4,106.8,559345,526197,6512,0.4012,21,16
Claude Code,Chrome DevTools,MCP,Windows,C1,78.8,592301,555708,6171,0.4175,20,14
Claude Code,Chrome DevTools,MCP,Windows,C2,111.9,482381,446017,5828,0.3875,17,15
Claude Code,Chrome DevTools,MCP,Windows,C3,69,488951,450930,5696,0.3942,17,17
Claude Code,Chrome DevTools,MCP,Windows,C4,73.2,528741,488749,5632,0.4103,18,16
Codex,agent-browser,CLI + skill,macOS,C1,90.1,502096,467456,2210,1.0004,16,27
Codex,agent-browser,CLI + skill,macOS,C2,95.3,358084,327424,2719,0.8,15,27
Codex,agent-browser,CLI + skill,macOS,C3,94.4,427258,384000,2625,0.9478,14,27
Codex,agent-browser,CLI + skill,macOS,C4,83.5,478797,446720,2286,0.9578,16,27
Codex,agent-browser,CLI + skill,WSL,C1,78.8,394667,359296,2313,0.8587,14,25
Codex,agent-browser,CLI + skill,WSL,C2,70,278405,254080,2046,0.6756,12,23
Codex,agent-browser,CLI + skill,WSL,C3,75.4,396641,366336,2190,0.8549,14,23
Codex,agent-browser,CLI + skill,WSL,C4,126.8,424608,390784,2105,0.9103,14,23
Codex,agent-browser,CLI + skill,Windows,C1,92.3,419482,376448,1802,0.8969,13,26
Codex,agent-browser,CLI + skill,Windows,C2,106.2,437748,396928,2126,0.9114,14,29
Codex,agent-browser,CLI + skill,Windows,C3,104.8,478086,436736,1856,0.943,15,29
Codex,agent-browser,CLI + skill,Windows,C4,90.6,505292,460416,1616,0.99,15,28
Codex,agent-browser,"CLI, no skill",macOS,C1,90.5,290195,268800,2401,0.6788,14,22
Codex,agent-browser,"CLI, no skill",macOS,C2,67.6,281581,261760,1857,0.6288,14,24
Codex,agent-browser,"CLI, no skill",macOS,C3,76.8,334785,315136,1883,0.6818,17,22
Codex,agent-browser,"CLI, no skill",macOS,C4,75.9,317529,297088,1875,0.6712,16,25
Codex,agent-browser,"CLI, no skill",WSL,C1,59.3,227881,203904,1904,0.5389,13,20
Codex,agent-browser,"CLI, no skill",WSL,C2,78.2,287027,232704,1905,0.8712,14,19
Codex,agent-browser,"CLI, no skill",WSL,C3,59.2,265128,215040,1641,0.798,14,18
Codex,agent-browser,"CLI, no skill",WSL,C4,68.6,256473,235904,1871,0.6111,13,20
Codex,agent-browser,"CLI, no skill",Windows,C1,90,291446,259712,2273,0.6907,14,23
Codex,agent-browser,"CLI, no skill",Windows,C2,87.1,272835,254336,1867,0.611,14,23
Codex,agent-browser,"CLI, no skill",Windows,C3,107,318932,297600,2269,0.7016,16,27
Codex,agent-browser,"CLI, no skill",Windows,C4,95.3,308420,268800,1644,0.7472,15,24
Codex,agent-browser,MCP,macOS,C1,86.3,473068,438272,2099,0.9811,14,21
Codex,agent-browser,MCP,macOS,C2,76.4,411651,364160,2080,0.9731,13,23
Codex,agent-browser,MCP,macOS,C3,77,408954,376576,1839,0.8683,13,20
Codex,agent-browser,MCP,macOS,C4,66.8,417015,384000,1979,0.8891,13,20
Codex,agent-browser,MCP,WSL,C1,67.8,336157,304768,1821,0.7857,11,21
Codex,agent-browser,MCP,WSL,C2,60.9,334115,303232,1755,0.7758,11,20
Codex,agent-browser,MCP,WSL,C3,62.4,341903,308096,1790,0.8117,11,20
Codex,agent-browser,MCP,WSL,C4,63.5,332658,301824,1636,0.768,11,18
Codex,agent-browser,MCP,Windows,C1,81.7,425217,391424,1838,0.8985,13,22
Codex,agent-browser,MCP,Windows,C2,77.2,429452,393856,1799,0.917,13,21
Codex,agent-browser,MCP,Windows,C3,80.2,463736,429824,2022,0.9472,14,20
Codex,agent-browser,MCP,Windows,C4,59.2,417455,385792,1778,0.8823,13,22
Codex,Playwright,CLI + skill,macOS,C1,106.4,393981,362496,2526,0.8796,18,20
Codex,Playwright,CLI + skill,macOS,C2,91.6,282923,241536,2243,0.8436,15,20
Codex,Playwright,CLI + skill,macOS,C3,93.9,329536,309248,2227,0.6995,16,19
Codex,Playwright,CLI + skill,macOS,C4,70.2,331614,311424,1903,0.6845,16,17
Codex,Playwright,CLI + skill,WSL,C1,103.9,330372,289920,2074,0.8741,16,16
Codex,Playwright,CLI + skill,WSL,C2,99.3,323012,280448,2672,0.9157,16,15
Codex,Playwright,CLI + skill,WSL,C3,90.1,312307,269312,2176,0.8841,16,19
Codex,Playwright,CLI + skill,WSL,C4,78.2,210642,183296,1975,0.6315,12,16
Codex,Playwright,CLI + skill,Windows,C1,137.1,290886,263808,2209,0.645,14,14
Codex,Playwright,CLI + skill,Windows,C2,150.7,399080,370944,2227,0.7637,19,14
Codex,Playwright,CLI + skill,Windows,C3,126.6,306830,258944,1801,0.8279,15,15
Codex,Playwright,CLI + skill,Windows,C4,128.3,291876,273536,1911,0.6297,14,15
Codex,Playwright,"CLI, no skill",macOS,C1,84.3,223557,205056,2130,0.5726,12,17
Codex,Playwright,"CLI, no skill",macOS,C2,73.9,233909,206720,2003,0.6548,14,16
Codex,Playwright,"CLI, no skill",macOS,C3,97.2,247066,230144,2337,0.5922,15,15
Codex,Playwright,"CLI, no skill",macOS,C4,105.9,293966,257536,2239,0.8098,17,19
Codex,Playwright,"CLI, no skill",WSL,C1,81.5,246482,230272,2388,0.5878,14,18
Codex,Playwright,"CLI, no skill",WSL,C2,80.7,223968,193408,2097,0.6799,14,21
Codex,Playwright,"CLI, no skill",WSL,C3,79.1,230884,196864,2027,0.6684,14,18
Codex,Playwright,"CLI, no skill",WSL,C4,81.9,239769,223616,2171,0.5697,14,14
Codex,Playwright,"CLI, no skill",Windows,C1,136,306016,278016,2518,0.6839,16,10
Codex,Playwright,"CLI, no skill",Windows,C2,124.9,209276,185088,2304,0.6205,13,13
Codex,Playwright,"CLI, no skill",Windows,C3,160.3,367900,332544,2399,0.8833,20,20
Codex,Playwright,"CLI, no skill",Windows,C4,152.6,333621,314624,2174,0.6905,18,21
Codex,Playwright,MCP,macOS,C1,116.5,334699,310784,2215,0.7367,15,18
Codex,Playwright,MCP,macOS,C2,113.1,300783,274816,2066,0.7138,16,15
Codex,Playwright,MCP,macOS,C3,71.7,281815,264448,1609,0.5946,15,19
Codex,Playwright,MCP,macOS,C4,66.1,241582,223872,1848,0.5694,13,17
Codex,Playwright,MCP,WSL,C1,89.4,283805,260480,2095,0.6285,15,19
Codex,Playwright,MCP,WSL,C2,79.1,244789,221312,1975,0.5848,13,18
Codex,Playwright,MCP,WSL,C3,83.6,239171,211712,2113,0.668,13,17
Codex,Playwright,MCP,WSL,C4,70.1,265970,247296,1984,0.6092,14,19
Codex,Playwright,MCP,Windows,C1,96.6,227696,200832,2198,0.5794,12,16
Codex,Playwright,MCP,Windows,C2,111.2,276835,257408,2512,0.6556,14,14
Codex,Playwright,MCP,Windows,C3,89.6,255920,234496,2071,0.6295,12,19
Codex,Playwright,MCP,Windows,C4,87.3,188200,160128,2189,0.5503,10,14
Codex,Chrome DevTools,CLI + skill,macOS,C1,67.9,186379,169088,1949,0.5154,10,16
Codex,Chrome DevTools,CLI + skill,macOS,C2,68.9,200361,178176,1807,0.5204,11,17
Codex,Chrome DevTools,CLI + skill,macOS,C3,73.2,184909,160000,1820,0.5301,10,17
Codex,Chrome DevTools,CLI + skill,macOS,C4,70.4,220817,199040,1967,0.5544,12,17
Codex,Chrome DevTools,CLI + skill,WSL,C1,76,217837,183936,1820,0.6899,12,17
Codex,Chrome DevTools,CLI + skill,WSL,C2,74.4,194059,180736,1849,0.4824,11,17
Codex,Chrome DevTools,CLI + skill,WSL,C3,78.4,257037,238848,1956,0.5945,14,17
Codex,Chrome DevTools,CLI + skill,WSL,C4,72.1,193834,177408,1837,0.5095,11,17
Codex,Chrome DevTools,CLI + skill,Windows,C1,139.9,232539,206592,1916,0.5619,12,17
Codex,Chrome DevTools,CLI + skill,Windows,C2,142.5,172584,150272,1708,0.4588,10,16
Codex,Chrome DevTools,CLI + skill,Windows,C3,141.9,189391,171392,2007,0.5289,10,17
Codex,Chrome DevTools,CLI + skill,Windows,C4,167.8,222617,207872,1925,0.5288,13,17
Codex,Chrome DevTools,"CLI, no skill",macOS,C1,73.5,207489,190720,1804,0.5246,12,17
Codex,Chrome DevTools,"CLI, no skill",macOS,C2,87.4,283141,264320,2120,0.6345,15,17
Codex,Chrome DevTools,"CLI, no skill",macOS,C3,85.2,234010,213120,1891,0.5926,13,17
Codex,Chrome DevTools,"CLI, no skill",macOS,C4,70.7,226207,198528,1795,0.6411,13,17
Codex,Chrome DevTools,"CLI, no skill",WSL,C1,107.2,233125,215808,1996,0.5648,13,18
Codex,Chrome DevTools,"CLI, no skill",WSL,C2,68.8,227191,210432,1855,0.5468,13,18
Codex,Chrome DevTools,"CLI, no skill",WSL,C3,79.5,232221,214656,1953,0.564,13,18
Codex,Chrome DevTools,"CLI, no skill",WSL,C4,73.5,232059,214784,1860,0.5565,13,18
Codex,Chrome DevTools,"CLI, no skill",Windows,C1,144.7,239154,212608,1965,0.5763,13,18
Codex,Chrome DevTools,"CLI, no skill",Windows,C2,138.1,239063,202368,1910,0.6648,13,17
Codex,Chrome DevTools,"CLI, no skill",Windows,C3,168.5,263123,214016,2111,0.8106,14,18
Codex,Chrome DevTools,"CLI, no skill",Windows,C4,276.5,231905,190080,2003,0.7995,13,18
Codex,Chrome DevTools,MCP,macOS,C1,59.9,165715,136960,1956,0.5983,11,14
Codex,Chrome DevTools,MCP,macOS,C2,81.7,339298,298496,2113,0.8882,17,15
Codex,Chrome DevTools,MCP,macOS,C3,64.1,212156,195584,1931,0.5466,11,14
Codex,Chrome DevTools,MCP,macOS,C4,88.5,209419,193536,2075,0.5321,12,15
Codex,Chrome DevTools,MCP,WSL,C1,80.6,257717,236288,2079,0.6305,17,14
Codex,Chrome DevTools,MCP,WSL,C2,62.5,247106,221568,1839,0.6449,12,14
Codex,Chrome DevTools,MCP,WSL,C3,66.5,173341,151296,1872,0.4653,11,14
Codex,Chrome DevTools,MCP,WSL,C4,66.2,247873,228736,1854,0.5888,12,15
Codex,Chrome DevTools,MCP,Windows,C1,93,365295,336640,1748,0.7106,17,14
Codex,Chrome DevTools,MCP,Windows,C2,112.2,241473,215296,2122,0.5832,12,15
Codex,Chrome DevTools,MCP,Windows,C3,88.3,233948,213248,2093,0.6021,11,15
Codex,Chrome DevTools,MCP,Windows,C4,84.2,189917,158080,2064,0.6568,11,15
```

**Show the 216 runs**

**Every run**

| OS | Run | Time (s) | Input tokens | Cached input | Output tokens | Cost (USD) | Model calls | Browser commands |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Claude Code, agent-browser, CLI + skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 116.8 | 623,265 | 584,110 | 9,156 | $0.50 | 23 | 55 |
| macOS | C2 | 97.4 | 569,199 | 534,281 | 5,713 | $0.42 | 19 | 39 |
| macOS | C3 | 142.4 | 682,163 | 646,584 | 9,044 | $0.52 | 23 | 39 |
| macOS | C4 | 132.3 | 1,093,881 | 1,048,684 | 10,628 | $0.68 | 29 | 52 |
| WSL | C1 | 95.7 | 471,985 | 437,935 | 6,209 | $0.41 | 17 | 36 |
| WSL | C2 | 82.2 | 543,769 | 509,800 | 6,843 | $0.44 | 18 | 41 |
| WSL | C3 | 83.9 | 486,134 | 452,068 | 6,763 | $0.42 | 17 | 24 |
| WSL | C4 | 95.8 | 734,957 | 691,765 | 6,075 | $0.50 | 20 | 28 |
| Windows | C1 | 129.2 | 681,670 | 646,381 | 6,191 | $0.46 | 22 | 43 |
| Windows | C2 | 109.9 | 706,145 | 668,782 | 8,985 | $0.54 | 22 | 52 |
| Windows | C3 | 92.7 | 594,483 | 562,471 | 6,643 | $0.43 | 21 | 46 |
| Windows | C4 | 119.6 | 962,950 | 917,408 | 9,827 | $0.64 | 25 | 35 |
| Claude Code, agent-browser, CLI, no skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 101.6 | 341,587 | 321,506 | 7,700 | $0.35 | 20 | 42 |
| macOS | C2 | 88.9 | 304,730 | 288,599 | 5,877 | $0.28 | 19 | 44 |
| macOS | C3 | 99.9 | 346,352 | 329,634 | 7,201 | $0.32 | 22 | 41 |
| macOS | C4 | 103.0 | 340,204 | 323,655 | 5,568 | $0.29 | 21 | 38 |
| WSL | C1 | 91.5 | 269,273 | 254,068 | 5,683 | $0.27 | 18 | 34 |
| WSL | C2 | 73.0 | 244,496 | 229,518 | 5,356 | $0.26 | 16 | 35 |
| WSL | C3 | 91.3 | 360,465 | 343,898 | 7,731 | $0.33 | 23 | 33 |
| WSL | C4 | 76.0 | 294,146 | 276,951 | 6,077 | $0.29 | 18 | 34 |
| Windows | C1 | 106.8 | 395,009 | 377,081 | 7,175 | $0.34 | 23 | 40 |
| Windows | C2 | 83.0 | 304,086 | 290,226 | 5,861 | $0.28 | 19 | 40 |
| Windows | C3 | 126.1 | 301,055 | 287,879 | 5,877 | $0.28 | 20 | 37 |
| Windows | C4 | 121.8 | 270,652 | 256,583 | 5,848 | $0.27 | 19 | 37 |
| Claude Code, agent-browser, MCP |   |   |   |   |   |   |   |   |
| macOS | C1 | 105.4 | 1,619,531 | 1,495,600 | 8,756 | $1.09 | 21 | 27 |
| macOS | C2 | 107.2 | 1,403,413 | 1,277,608 | 8,323 | $1.05 | 19 | 27 |
| macOS | C3 | 101.6 | 1,545,504 | 1,438,812 | 9,155 | $1.00 | 21 | 28 |
| macOS | C4 | 97.0 | 1,603,418 | 1,489,693 | 8,623 | $1.04 | 20 | 28 |
| WSL | C1 | 130.9 | 1,617,843 | 1,514,370 | 8,517 | $0.99 | 23 | 21 |
| WSL | C2 | 95.9 | 1,498,581 | 1,394,831 | 7,947 | $0.96 | 21 | 26 |
| WSL | C3 | 86.7 | 1,289,803 | 1,208,208 | 8,102 | $0.81 | 20 | 20 |
| WSL | C4 | 101.5 | 1,577,932 | 1,492,459 | 9,168 | $0.91 | 23 | 25 |
| Windows | C1 | 133.1 | 1,426,308 | 1,357,518 | 9,130 | $0.98 | 20 | 30 |
| Windows | C2 | 168.2 | 1,904,787 | 1,792,941 | 8,486 | $1.09 | 24 | 25 |
| Windows | C3 | 143.8 | 1,864,586 | 1,798,407 | 8,950 | $1.05 | 26 | 26 |
| Windows | C4 | 169.6 | 1,992,230 | 1,881,767 | 10,233 | $1.13 | 27 | 27 |
| Claude Code, Playwright, CLI + skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 127.5 | 399,690 | 375,665 | 6,560 | $0.35 | 18 | 28 |
| macOS | C2 | 77.5 | 304,364 | 278,452 | 5,187 | $0.29 | 15 | 23 |
| macOS | C3 | 90.9 | 316,991 | 297,255 | 5,279 | $0.29 | 16 | 22 |
| macOS | C4 | 78.8 | 328,600 | 307,213 | 6,040 | $0.32 | 16 | 18 |
| WSL | C1 | 84.5 | 363,244 | 339,728 | 6,221 | $0.33 | 17 | 23 |
| WSL | C2 | 101.6 | 381,186 | 356,266 | 5,429 | $0.33 | 16 | 20 |
| WSL | C3 | 76.6 | 314,550 | 294,746 | 5,618 | $0.30 | 16 | 20 |
| WSL | C4 | 74.0 | 256,717 | 236,700 | 4,776 | $0.27 | 13 | 20 |
| Windows | C1 | 99.5 | 393,098 | 365,680 | 6,947 | $0.37 | 16 | 22 |
| Windows | C2 | 89.3 | 282,898 | 261,932 | 4,714 | $0.28 | 14 | 20 |
| Windows | C3 | 94.3 | 335,328 | 313,174 | 6,791 | $0.34 | 16 | 22 |
| Windows | C4 | 100.0 | 379,986 | 357,800 | 7,002 | $0.35 | 18 | 22 |
| Claude Code, Playwright, CLI, no skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 97.2 | 410,283 | 386,802 | 6,966 | $0.36 | 21 | 30 |
| macOS | C2 | 85.4 | 251,297 | 231,793 | 6,492 | $0.30 | 15 | 18 |
| macOS | C3 | 102.1 | 374,628 | 353,625 | 7,407 | $0.35 | 21 | 25 |
| macOS | C4 | 79.9 | 292,361 | 273,627 | 5,868 | $0.29 | 17 | 21 |
| WSL | C1 | 93.7 | 319,087 | 299,570 | 6,488 | $0.31 | 18 | 22 |
| WSL | C2 | 95.2 | 325,868 | 308,670 | 7,380 | $0.33 | 19 | 22 |
| WSL | C3 | 116.4 | 365,122 | 344,874 | 6,989 | $0.34 | 20 | 24 |
| WSL | C4 | 97.1 | 341,101 | 322,237 | 5,763 | $0.30 | 20 | 25 |
| Windows | C1 | 107.6 | 387,138 | 364,969 | 7,251 | $0.35 | 21 | 27 |
| Windows | C2 | 125.1 | 316,374 | 300,451 | 5,270 | $0.28 | 19 | 25 |
| Windows | C3 | 101.2 | 373,167 | 352,375 | 6,693 | $0.34 | 20 | 21 |
| Windows | C4 | 115.9 | 384,501 | 364,510 | 6,187 | $0.32 | 22 | 26 |
| Claude Code, Playwright, MCP |   |   |   |   |   |   |   |   |
| macOS | C1 | 108.4 | 521,899 | 487,825 | 7,236 | $0.41 | 21 | 17 |
| macOS | C2 | 112.4 | 506,912 | 472,458 | 7,584 | $0.42 | 20 | 17 |
| macOS | C3 | 90.9 | 528,905 | 495,588 | 7,384 | $0.41 | 22 | 13 |
| macOS | C4 | 118.5 | 655,019 | 621,922 | 7,202 | $0.43 | 27 | 13 |
| WSL | C1 | 86.7 | 518,440 | 499,450 | 7,166 | $0.41 | 21 | 15 |
| WSL | C2 | 93.1 | 417,396 | 386,804 | 6,597 | $0.36 | 19 | 17 |
| WSL | C3 | 115.5 | 690,872 | 665,007 | 9,646 | $0.52 | 27 | 19 |
| WSL | C4 | 93.3 | 449,468 | 416,054 | 7,157 | $0.39 | 18 | 15 |
| Windows | C1 | 102.9 | 580,763 | 546,830 | 7,959 | $0.44 | 23 | 15 |
| Windows | C2 | 91.3 | 463,217 | 430,293 | 7,551 | $0.40 | 19 | 16 |
| Windows | C3 | 93.3 | 455,884 | 423,204 | 7,787 | $0.40 | 19 | 15 |
| Windows | C4 | 104.6 | 481,955 | 449,264 | 6,662 | $0.39 | 20 | 15 |
| Claude Code, Chrome DevTools, CLI + skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 78.9 | 332,835 | 307,057 | 6,041 | $0.31 | 17 | 21 |
| macOS | C2 | 79.1 | 326,953 | 306,648 | 5,782 | $0.31 | 17 | 23 |
| macOS | C3 | 78.9 | 348,502 | 326,433 | 6,102 | $0.33 | 17 | 21 |
| macOS | C4 | 111.8 | 291,502 | 271,629 | 5,977 | $0.30 | 15 | 18 |
| WSL | C1 | 118.9 | 508,343 | 482,961 | 9,068 | $0.43 | 23 | 30 |
| WSL | C2 | 94.2 | 312,100 | 292,979 | 4,667 | $0.28 | 16 | 19 |
| WSL | C3 | 78.7 | 283,404 | 264,443 | 5,767 | $0.29 | 15 | 20 |
| WSL | C4 | 85.0 | 378,445 | 357,446 | 6,385 | $0.33 | 19 | 22 |
| Windows | C1 | 116.2 | 355,723 | 335,232 | 6,265 | $0.32 | 18 | 22 |
| Windows | C2 | 119.8 | 367,891 | 348,403 | 5,998 | $0.32 | 19 | 21 |
| Windows | C3 | 100.3 | 328,594 | 308,513 | 6,534 | $0.32 | 17 | 22 |
| Windows | C4 | 117.8 | 498,954 | 457,626 | 6,890 | $0.46 | 18 | 25 |
| Claude Code, Chrome DevTools, CLI, no skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 97.7 | 345,847 | 326,196 | 5,925 | $0.31 | 19 | 24 |
| macOS | C2 | 84.3 | 387,549 | 366,560 | 6,166 | $0.34 | 19 | 20 |
| macOS | C3 | 83.9 | 304,904 | 286,439 | 6,221 | $0.30 | 18 | 24 |
| macOS | C4 | 89.2 | 325,593 | 304,468 | 5,299 | $0.30 | 17 | 23 |
| WSL | C1 | 97.3 | 342,741 | 323,253 | 5,392 | $0.29 | 19 | 22 |
| WSL | C2 | 110.1 | 401,020 | 380,686 | 7,951 | $0.36 | 22 | 24 |
| WSL | C3 | 84.9 | 373,668 | 350,745 | 5,923 | $0.33 | 18 | 24 |
| WSL | C4 | 80.9 | 389,920 | 367,817 | 5,633 | $0.32 | 19 | 24 |
| Windows | C1 | 102.3 | 346,841 | 326,638 | 5,875 | $0.31 | 18 | 24 |
| Windows | C2 | 110.9 | 288,332 | 271,164 | 5,048 | $0.27 | 17 | 19 |
| Windows | C3 | 115.8 | 357,968 | 336,528 | 6,806 | $0.33 | 19 | 23 |
| Windows | C4 | 105.6 | 349,957 | 329,889 | 6,637 | $0.33 | 19 | 21 |
| Claude Code, Chrome DevTools, MCP |   |   |   |   |   |   |   |   |
| macOS | C1 | 64.2 | 518,779 | 482,736 | 5,147 | $0.38 | 18 | 13 |
| macOS | C2 | 84.6 | 514,662 | 476,908 | 5,675 | $0.40 | 18 | 14 |
| macOS | C3 | 66.3 | 467,466 | 429,759 | 5,553 | $0.39 | 16 | 15 |
| macOS | C4 | 69.8 | 454,948 | 418,393 | 5,455 | $0.38 | 16 | 14 |
| WSL | C1 | 61.9 | 461,992 | 425,869 | 5,078 | $0.37 | 16 | 14 |
| WSL | C2 | 83.8 | 590,591 | 557,451 | 6,481 | $0.41 | 22 | 18 |
| WSL | C3 | 69.2 | 535,587 | 495,244 | 5,490 | $0.41 | 18 | 15 |
| WSL | C4 | 106.8 | 559,345 | 526,197 | 6,512 | $0.40 | 21 | 16 |
| Windows | C1 | 78.8 | 592,301 | 555,708 | 6,171 | $0.42 | 20 | 14 |
| Windows | C2 | 111.9 | 482,381 | 446,017 | 5,828 | $0.39 | 17 | 15 |
| Windows | C3 | 69.0 | 488,951 | 450,930 | 5,696 | $0.39 | 17 | 17 |
| Windows | C4 | 73.2 | 528,741 | 488,749 | 5,632 | $0.41 | 18 | 16 |
| Codex, agent-browser, CLI + skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 90.1 | 502,096 | 467,456 | 2,210 | $1.00 | 16 | 27 |
| macOS | C2 | 95.3 | 358,084 | 327,424 | 2,719 | $0.80 | 15 | 27 |
| macOS | C3 | 94.4 | 427,258 | 384,000 | 2,625 | $0.95 | 14 | 27 |
| macOS | C4 | 83.5 | 478,797 | 446,720 | 2,286 | $0.96 | 16 | 27 |
| WSL | C1 | 78.8 | 394,667 | 359,296 | 2,313 | $0.86 | 14 | 25 |
| WSL | C2 | 70.0 | 278,405 | 254,080 | 2,046 | $0.68 | 12 | 23 |
| WSL | C3 | 75.4 | 396,641 | 366,336 | 2,190 | $0.85 | 14 | 23 |
| WSL | C4 | 126.8 | 424,608 | 390,784 | 2,105 | $0.91 | 14 | 23 |
| Windows | C1 | 92.3 | 419,482 | 376,448 | 1,802 | $0.90 | 13 | 26 |
| Windows | C2 | 106.2 | 437,748 | 396,928 | 2,126 | $0.91 | 14 | 29 |
| Windows | C3 | 104.8 | 478,086 | 436,736 | 1,856 | $0.94 | 15 | 29 |
| Windows | C4 | 90.6 | 505,292 | 460,416 | 1,616 | $0.99 | 15 | 28 |
| Codex, agent-browser, CLI, no skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 90.5 | 290,195 | 268,800 | 2,401 | $0.68 | 14 | 22 |
| macOS | C2 | 67.6 | 281,581 | 261,760 | 1,857 | $0.63 | 14 | 24 |
| macOS | C3 | 76.8 | 334,785 | 315,136 | 1,883 | $0.68 | 17 | 22 |
| macOS | C4 | 75.9 | 317,529 | 297,088 | 1,875 | $0.67 | 16 | 25 |
| WSL | C1 | 59.3 | 227,881 | 203,904 | 1,904 | $0.54 | 13 | 20 |
| WSL | C2 | 78.2 | 287,027 | 232,704 | 1,905 | $0.87 | 14 | 19 |
| WSL | C3 | 59.2 | 265,128 | 215,040 | 1,641 | $0.80 | 14 | 18 |
| WSL | C4 | 68.6 | 256,473 | 235,904 | 1,871 | $0.61 | 13 | 20 |
| Windows | C1 | 90.0 | 291,446 | 259,712 | 2,273 | $0.69 | 14 | 23 |
| Windows | C2 | 87.1 | 272,835 | 254,336 | 1,867 | $0.61 | 14 | 23 |
| Windows | C3 | 107.0 | 318,932 | 297,600 | 2,269 | $0.70 | 16 | 27 |
| Windows | C4 | 95.3 | 308,420 | 268,800 | 1,644 | $0.75 | 15 | 24 |
| Codex, agent-browser, MCP |   |   |   |   |   |   |   |   |
| macOS | C1 | 86.3 | 473,068 | 438,272 | 2,099 | $0.98 | 14 | 21 |
| macOS | C2 | 76.4 | 411,651 | 364,160 | 2,080 | $0.97 | 13 | 23 |
| macOS | C3 | 77.0 | 408,954 | 376,576 | 1,839 | $0.87 | 13 | 20 |
| macOS | C4 | 66.8 | 417,015 | 384,000 | 1,979 | $0.89 | 13 | 20 |
| WSL | C1 | 67.8 | 336,157 | 304,768 | 1,821 | $0.79 | 11 | 21 |
| WSL | C2 | 60.9 | 334,115 | 303,232 | 1,755 | $0.78 | 11 | 20 |
| WSL | C3 | 62.4 | 341,903 | 308,096 | 1,790 | $0.81 | 11 | 20 |
| WSL | C4 | 63.5 | 332,658 | 301,824 | 1,636 | $0.77 | 11 | 18 |
| Windows | C1 | 81.7 | 425,217 | 391,424 | 1,838 | $0.90 | 13 | 22 |
| Windows | C2 | 77.2 | 429,452 | 393,856 | 1,799 | $0.92 | 13 | 21 |
| Windows | C3 | 80.2 | 463,736 | 429,824 | 2,022 | $0.95 | 14 | 20 |
| Windows | C4 | 59.2 | 417,455 | 385,792 | 1,778 | $0.88 | 13 | 22 |
| Codex, Playwright, CLI + skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 106.4 | 393,981 | 362,496 | 2,526 | $0.88 | 18 | 20 |
| macOS | C2 | 91.6 | 282,923 | 241,536 | 2,243 | $0.84 | 15 | 20 |
| macOS | C3 | 93.9 | 329,536 | 309,248 | 2,227 | $0.70 | 16 | 19 |
| macOS | C4 | 70.2 | 331,614 | 311,424 | 1,903 | $0.68 | 16 | 17 |
| WSL | C1 | 103.9 | 330,372 | 289,920 | 2,074 | $0.87 | 16 | 16 |
| WSL | C2 | 99.3 | 323,012 | 280,448 | 2,672 | $0.92 | 16 | 15 |
| WSL | C3 | 90.1 | 312,307 | 269,312 | 2,176 | $0.88 | 16 | 19 |
| WSL | C4 | 78.2 | 210,642 | 183,296 | 1,975 | $0.63 | 12 | 16 |
| Windows | C1 | 137.1 | 290,886 | 263,808 | 2,209 | $0.65 | 14 | 14 |
| Windows | C2 | 150.7 | 399,080 | 370,944 | 2,227 | $0.76 | 19 | 14 |
| Windows | C3 | 126.6 | 306,830 | 258,944 | 1,801 | $0.83 | 15 | 15 |
| Windows | C4 | 128.3 | 291,876 | 273,536 | 1,911 | $0.63 | 14 | 15 |
| Codex, Playwright, CLI, no skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 84.3 | 223,557 | 205,056 | 2,130 | $0.57 | 12 | 17 |
| macOS | C2 | 73.9 | 233,909 | 206,720 | 2,003 | $0.65 | 14 | 16 |
| macOS | C3 | 97.2 | 247,066 | 230,144 | 2,337 | $0.59 | 15 | 15 |
| macOS | C4 | 105.9 | 293,966 | 257,536 | 2,239 | $0.81 | 17 | 19 |
| WSL | C1 | 81.5 | 246,482 | 230,272 | 2,388 | $0.59 | 14 | 18 |
| WSL | C2 | 80.7 | 223,968 | 193,408 | 2,097 | $0.68 | 14 | 21 |
| WSL | C3 | 79.1 | 230,884 | 196,864 | 2,027 | $0.67 | 14 | 18 |
| WSL | C4 | 81.9 | 239,769 | 223,616 | 2,171 | $0.57 | 14 | 14 |
| Windows | C1 | 136.0 | 306,016 | 278,016 | 2,518 | $0.68 | 16 | 10 |
| Windows | C2 | 124.9 | 209,276 | 185,088 | 2,304 | $0.62 | 13 | 13 |
| Windows | C3 | 160.3 | 367,900 | 332,544 | 2,399 | $0.88 | 20 | 20 |
| Windows | C4 | 152.6 | 333,621 | 314,624 | 2,174 | $0.69 | 18 | 21 |
| Codex, Playwright, MCP |   |   |   |   |   |   |   |   |
| macOS | C1 | 116.5 | 334,699 | 310,784 | 2,215 | $0.74 | 15 | 18 |
| macOS | C2 | 113.1 | 300,783 | 274,816 | 2,066 | $0.71 | 16 | 15 |
| macOS | C3 | 71.7 | 281,815 | 264,448 | 1,609 | $0.59 | 15 | 19 |
| macOS | C4 | 66.1 | 241,582 | 223,872 | 1,848 | $0.57 | 13 | 17 |
| WSL | C1 | 89.4 | 283,805 | 260,480 | 2,095 | $0.63 | 15 | 19 |
| WSL | C2 | 79.1 | 244,789 | 221,312 | 1,975 | $0.58 | 13 | 18 |
| WSL | C3 | 83.6 | 239,171 | 211,712 | 2,113 | $0.67 | 13 | 17 |
| WSL | C4 | 70.1 | 265,970 | 247,296 | 1,984 | $0.61 | 14 | 19 |
| Windows | C1 | 96.6 | 227,696 | 200,832 | 2,198 | $0.58 | 12 | 16 |
| Windows | C2 | 111.2 | 276,835 | 257,408 | 2,512 | $0.66 | 14 | 14 |
| Windows | C3 | 89.6 | 255,920 | 234,496 | 2,071 | $0.63 | 12 | 19 |
| Windows | C4 | 87.3 | 188,200 | 160,128 | 2,189 | $0.55 | 10 | 14 |
| Codex, Chrome DevTools, CLI + skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 67.9 | 186,379 | 169,088 | 1,949 | $0.52 | 10 | 16 |
| macOS | C2 | 68.9 | 200,361 | 178,176 | 1,807 | $0.52 | 11 | 17 |
| macOS | C3 | 73.2 | 184,909 | 160,000 | 1,820 | $0.53 | 10 | 17 |
| macOS | C4 | 70.4 | 220,817 | 199,040 | 1,967 | $0.55 | 12 | 17 |
| WSL | C1 | 76.0 | 217,837 | 183,936 | 1,820 | $0.69 | 12 | 17 |
| WSL | C2 | 74.4 | 194,059 | 180,736 | 1,849 | $0.48 | 11 | 17 |
| WSL | C3 | 78.4 | 257,037 | 238,848 | 1,956 | $0.59 | 14 | 17 |
| WSL | C4 | 72.1 | 193,834 | 177,408 | 1,837 | $0.51 | 11 | 17 |
| Windows | C1 | 139.9 | 232,539 | 206,592 | 1,916 | $0.56 | 12 | 17 |
| Windows | C2 | 142.5 | 172,584 | 150,272 | 1,708 | $0.46 | 10 | 16 |
| Windows | C3 | 141.9 | 189,391 | 171,392 | 2,007 | $0.53 | 10 | 17 |
| Windows | C4 | 167.8 | 222,617 | 207,872 | 1,925 | $0.53 | 13 | 17 |
| Codex, Chrome DevTools, CLI, no skill |   |   |   |   |   |   |   |   |
| macOS | C1 | 73.5 | 207,489 | 190,720 | 1,804 | $0.52 | 12 | 17 |
| macOS | C2 | 87.4 | 283,141 | 264,320 | 2,120 | $0.63 | 15 | 17 |
| macOS | C3 | 85.2 | 234,010 | 213,120 | 1,891 | $0.59 | 13 | 17 |
| macOS | C4 | 70.7 | 226,207 | 198,528 | 1,795 | $0.64 | 13 | 17 |
| WSL | C1 | 107.2 | 233,125 | 215,808 | 1,996 | $0.56 | 13 | 18 |
| WSL | C2 | 68.8 | 227,191 | 210,432 | 1,855 | $0.55 | 13 | 18 |
| WSL | C3 | 79.5 | 232,221 | 214,656 | 1,953 | $0.56 | 13 | 18 |
| WSL | C4 | 73.5 | 232,059 | 214,784 | 1,860 | $0.56 | 13 | 18 |
| Windows | C1 | 144.7 | 239,154 | 212,608 | 1,965 | $0.58 | 13 | 18 |
| Windows | C2 | 138.1 | 239,063 | 202,368 | 1,910 | $0.66 | 13 | 17 |
| Windows | C3 | 168.5 | 263,123 | 214,016 | 2,111 | $0.81 | 14 | 18 |
| Windows | C4 | 276.5 | 231,905 | 190,080 | 2,003 | $0.80 | 13 | 18 |
| Codex, Chrome DevTools, MCP |   |   |   |   |   |   |   |   |
| macOS | C1 | 59.9 | 165,715 | 136,960 | 1,956 | $0.60 | 11 | 14 |
| macOS | C2 | 81.7 | 339,298 | 298,496 | 2,113 | $0.89 | 17 | 15 |
| macOS | C3 | 64.1 | 212,156 | 195,584 | 1,931 | $0.55 | 11 | 14 |
| macOS | C4 | 88.5 | 209,419 | 193,536 | 2,075 | $0.53 | 12 | 15 |
| WSL | C1 | 80.6 | 257,717 | 236,288 | 2,079 | $0.63 | 17 | 14 |
| WSL | C2 | 62.5 | 247,106 | 221,568 | 1,839 | $0.64 | 12 | 14 |
| WSL | C3 | 66.5 | 173,341 | 151,296 | 1,872 | $0.47 | 11 | 14 |
| WSL | C4 | 66.2 | 247,873 | 228,736 | 1,854 | $0.59 | 12 | 15 |
| Windows | C1 | 93.0 | 365,295 | 336,640 | 1,748 | $0.71 | 17 | 14 |
| Windows | C2 | 112.2 | 241,473 | 215,296 | 2,122 | $0.58 | 12 | 15 |
| Windows | C3 | 88.3 | 233,948 | 213,248 | 2,093 | $0.60 | 11 | 15 |
| Windows | C4 | 84.2 | 189,917 | 158,080 | 2,064 | $0.66 | 11 | 15 |

Source: my study runs, 19 to 22 October 2026.

## Cite this study

Maximilian Fabian Kirchner (2026). Testing agent-browser, Playwright and Chrome DevTools with Claude Code and Codex. maximilianfabian.dev, published 1 November 2026. https://maximilianfabian.dev/blog/browser-tools-for-agents/

```bibtex
@misc{kirchner2026-browser-tools-for-agents,
  author       = {Kirchner, Maximilian Fabian},
  title        = {{Testing agent-browser, Playwright and Chrome DevTools with Claude Code and Codex}},
  howpublished = {\url{https://maximilianfabian.dev/blog/browser-tools-for-agents/}},
  year         = {2026},
  month        = nov,
  note         = {Published 1 November 2026},
}
```
