X · @teortaxesTex
· X / Twitter
> While the new DeepSeek V4 Flash performed similarly to the preview checkpoint in our agentic coding harness, it's dramatically stronger in one-shot,…
> While the new DeepSeek V4 Flash performed similarly to the preview checkpoint in our agentic coding harness, it's dramatically stronger in one-shot, fluid intelligenceUnexpected but I feel the sameIt’s NOT so much agent/general-harness-maxed as just smarterLeo Linsky: DeepSeek V4 Flash (07/31) results are now live on