ComposioBENCH
Try Bench

Harnesses

The agent around the model: tool loop, retry policy, context strategy, verifier. The same model performs differently across harnesses, so every harness gets its own ranking.

CompareEvery harness on one model, side by side
+2
8 harnesses
Every harness on the bench
  1. 01ClineDeepSeek V4 ProGPT-6 Astra
    79%
    Best time / task193s
  2. 02CodexDeepSeek V4 ProGPT-6 Astra
    76%
    Best time / task149s
  3. 03Hermes AgentDeepSeek V4 ProGPT-6 Astra
    76%
    Best time / task125s
  4. 04Command CodeDeepSeek V4 ProGPT-6 Astra
    76%
    Best time / task159s
  5. 05Claude CodeDeepSeek V4 ProGPT-6 Astra
    72%
    Best time / task198s
  6. 06OpenCodeDeepSeek V4 ProGPT-6 Astra
    69%
    Best time / task127s
  7. 07Pi AgentDeepSeek V4 ProGPT-6 Astra
    69%
    Best time / task128s
  8. 08DeepSeeklockedDeepSeek V4 Pro
    53%
    Best time / task388s

Ranked by models run, then best success · locked = vendor-locked to its own models · Composio Benchmark