
ClawBench
Benchmark scoring browser agents on 130 real tasks across live websites
- Pricing
- Open source
- Platforms
- CLI, Web
- GitHub stars
- 1.3K
- Last commit
- 12 days ago
- Language
- Python
- License
- Apache-2.0
About ClawBench
ClawBench is an open-source AI tool for browser automation. It is available on CLI and Web.
The benchmark gives agents everyday jobs such as booking flights, ordering food and applying for jobs, and checks if they submit the right thing. Scoring has two stages: the final HTTP request is intercepted and matched to the task schema, then an LLM judge compares the payload to the instruction.
V2 has 130 tasks over 63 platforms in 30 categories, V1 has 153 tasks. The leaderboard (as of 2026-05-20) shows claude-opus-4-7 on top at 54.6 percent intercepted for V2 with the hermes harness, and cost per task is reported.
To run it yourself: pip install clawbench-eval. Apache-2.0.
What it does
- Live websites
- Tasks run on real platforms.
- Two-stage scoring
- HTTP interception then LLM judge.
- V1 and V2 corpora
- 153 and 130 tasks.
- Leaderboard
- Reward, intercepted rate, cost per task.
- CLI runner
- pip install clawbench-eval.
Good for
- Compare models as browser agents
- Measure harness changes
- Reproduce leaderboard runs
Pricing
Open source
Apache-2.0
Prices change. Check claw-bench.com before you buy.
ClawBench FAQ
How is a task scored?
The final request is intercepted and then judged by an LLM.
How do I run it?
pip install clawbench-eval, then clawbench run.
Is ClawBench free?
Yes. ClawBench is open source, so you can use and self-host it for free.
What are the best alternatives to ClawBench?
The closest alternatives to ClawBench are GPT-3 Playground (OpenAI), ChatGPT (OpenAI), to teach, Heyday and Teach Anything.
Alternatives to ClawBench
-icon.png)
Free AI writing tool - Let the AI generate any text you can imagine
- Free
- Copywriting
-icon.jpg)
Ask any question or prompt to AI - The tool most other writing tools are based on
- Free
- Chat



