AnalysisAI ModelsAugust 9, 2026

DeepSeek V4 Flash 0731 tops Terminal-Bench 2.1 with 82.7%

DeepSeek V4 Flash 0731 leads Terminal-Bench 2.1 at 82.7% best accuracy across 89 tasks, with 368 of 445 trials passing. Every result links to its raw Harbor run and follows official TB 2.1 constraints: 5 trials per task, strict timeouts, and hardware limits.

2 sources

Daily brief

Get tomorrow's AI brief in your inbox

More stories today

Open the live feed