Request for official LexBench-Browser result review/submission

[Julia-Lex]

Hi Skyvern maintainers,

We maintain browseruse-agent-bench, a reproducible evaluation framework for browser agents. It includes LexBench-Browser, a public dataset with 210 no-login tasks across 107 real websites.

We already support running skyvern in the framework and would like to invite maintainers or power users to review the integration or submit an official result.

The goal is not to claim a ranking from our side, but to make the result reproducible and fair:

• confirm the recommended Skyvern version or commit
• record model provider and model ID
• record browser backend and launch mode
• include proxy or region notes if used
• include task-level artifacts
• disclose known skips, retries, provider incidents, or browser constraints
• allow maintainer review and rerun before listing as official

Useful links:

• Framework repo: https://github.com/lexmount/browseruse-agent-bench
• Dataset: https://huggingface.co/datasets/Lexmount/LexBench-Browser
• Evaluation protocol: https://github.com/lexmount/browseruse-agent-bench/blob/main/EVALUATION_PROTOCOL.md
• Result submission docs: https://github.com/lexmount/browseruse-agent-bench/tree/main/community/results

Would you be open to reviewing the integration or submitting an official Skyvern result? If there is a preferred configuration for benchmarking Skyvern, we would be happy to reflect that in the result metadata.

Thanks!

LexBench Team
lexbench@lexmount.com