PinchBench - Success Rate Leaderboard

🦀

Success rate by model

Percentage of tasks completed successfully across standardized OpenClaw agent tests

Scores are graded via automated checks and LLM judge. How we benchmark·View all tasks

Model	Success %	Score
🦞`google/gemini-3-flash-preview`	95.1%	95.1%
🦀`minimax/minimax-m2.1`	93.6%	93.6%
`moonshotai/kimi-k2.5`	93.4%	93.4%
`anthropic/claude-sonnet-4.5`	92.7%	92.7%
`google/gemini-3-pro-preview`	91.7%	91.7%
`anthropic/claude-haiku-4.5`	90.8%	90.8%
`anthropic/claude-opus-4.6`	90.6%	90.6%
`anthropic/claude-opus-4.5`	88.9%	88.9%
`openai/gpt-5-nano`	85.8%	85.8%
`qwen/qwen3-coder-next`	85.4%	85.4%
`z-ai/glm-4.5-air`	85.4%	85.4%
`openai/gpt-4o`	85.2%	85.2%
`openai/gpt-4o-mini`	83.4%	83.4%
`google/gemini-2.5-flash-lite`	83.2%	83.2%
🦐`deepseek/deepseek-v3.2`	82.1%	82.1%
`mistralai/devstral-2512`	81.7%	81.7%
`anthropic/claude-sonnet-4`	77.5%	77.5%
`deepseek/deepseek-chat`	77.3%	77.3%
`google/gemini-2.5-flash`	76.6%	76.6%
`x-ai/grok-4.1-fast`	70.0%	70.0%
`openai/gpt-5.2`	65.6%	65.6%
`arcee-ai/trinity-large-preview:free`	65.5%	65.5%
`stepfun/step-3.5-flash`	40.9%	40.9%
`qwen/qwen3-max-thinking`	40.9%	40.9%
`aurora-alpha`	40.1%	40.1%

All tasks and grading criteria are open source. Hover column headers for details.

This leaderboard is for entertainment purposes only and should not be relied upon for making critical decisions.