ComposioBENCH
Try Bench

DeepSeek V4 Pro

DeepSeek#2 of 2 by best run
Compare models
Best success60%in Command Code
Fastest run375sper task in Command Code
Best harness
Command Code
Graded runs
7 of 7 harnesses
Latest run
2026-09-04

Success rate

% of tasks passed by harness · higher is better
Command Code
60.0%
Hermes Agent
56.7%
Pi Agent
56.7%
Codex
53.3%
DeepSeek
53.3%
Claude Code
50.0%
OpenCode
50.0%

Time per task

Seconds per task by harness · lower is better
Command Code
375s
DeepSeek
388s
OpenCode
403s
Claude Code
414s
Hermes Agent
447s
Pi Agent
459s
Codex
482s

Runs

  1. 01Command Code
    60%
    Time / task375s
  2. 02Hermes Agent
    57%
    Time / task447s
  3. 03Pi Agent
    57%
    Time / task459s
  4. 04DeepSeek
    53%
    Time / task388s
  5. 05Codex
    53%
    Time / task482s
  6. 06OpenCode
    50%
    Time / task403s
  7. 07Claude Code
    50%
    Time / task414s