Leaderboard

The best model for coding agents right now, the cheapest one that still passes, and the fastest one that passes. Picked from tonight's measured results, not opinions.

Measured 2026-10-08 · 3 models tested through the public AnyModl API · How we test

Best for agent mode

deepseek/deepseek-chat

100% of agent tasks passed, p90 4.0 s

Tested 2026-10-08 · 6 tasks

Cheapest that passes

deepseek/deepseek-v4-flash

$0.59 per 1,000 passed tasks

Tested 2026-10-08 · 6 tasks

Fastest that passes

deepseek/deepseek-chat

p90 4.0 s per task (multi-turn included)

Tested 2026-10-08 · 6 tasks

“Passes” means at least 90% of the six harness tasks succeeded. Cost is at AnyModl customer prices per 1,000 passed tasks. Latency is the 90th percentile per task, including every turn of multi-step tool use. See the full scorecard.