Browser Agent Evals
readme.md →How today’s models perform on computer use benchmarks, compared on accuracy, cost, and speed.
15 runs, ranked by accuracy
accuracy per model
accuracy
project=stagehand-devbenchmark=hardbenchrows=15/83providers=4harnesses=6updated=2026-09-03T01:42:51Zrun=stagehand-evals
| claude-opus-5-5 | claude code | 1st89.0% | 326s | $0.510 | |
| gpt-6-astra | codex | 2nd87.0% | 320s | $3.073 | |
| gemini-4-argon | eve | 2nd87.0% | 525s | $1.422 | |
| gemini-4-argon | pi | 2nd87.0% | 630s | $1.938 | |
| claude-fable-5-1 | claude code | 84.0% | 552s | $1.631 | |
| gemini-4-argon | mastra | 84.0% | 657s | $2.127 | |
| gemini-4-argon | deep agents | 79.0% | 574s | $1.123 | |
| gpt-6.1-sol | codex | 79.0% | 2nd282s | 3rd$0.305 | |
| claude-sonnet-5-5 | claude code | 79.0% | 3rd288s | $0.324 | |
| claude-opus-5 | claude code | 78.0% | 678s | $2.437 | |
| gpt-6-sol | codex | 76.0% | 1st268s | $0.407 | |
| gpt-6-luna | codex | 74.0% | 311s | 1st$0.0225 | |
| claude-haiku-5-5 | claude code | 74.0% | 307s | 2nd$0.0679 | |
| gpt-5.6-sol | codex | 74.0% | 346s | $0.502 | |
| muse-wonder-1.0 | mastra | 74.0% | 374s | $4.210 |
- 1gemini-4-argon introductory rate through Nov 30, 2026, $2.843 after
- 2gemini-4-argon introductory rate through Nov 30, 2026, $3.876 after
- 3gemini-4-argon introductory rate through Nov 30, 2026, $4.254 after
- 4gemini-4-argon introductory rate through Nov 30, 2026, $2.246 after
Run your own evals
same harness, your models, your benchmark
Want to get your model or harness evaluated?
talk to an engineer →Your agent should be able to use a browser like you do.
The SDK for browser agents