🐳 DeepSeek Releases V4.1-Flash

On benchmarks, the model comes close to Kimi K3, GPT-5.6 Sol, and Claude Opus 5, and on some of them it comes out ahead. Pricing is $0.3/$1.2 per million input/output tokens, and during off-peak hours it drops by half again.

The model is multimodal, and in the API it replaces V4-Flash and V4-Flash-Vision-Exp. The Flash/Pro/Vision switch is gone from the web, it's now a single bot with a single window for text and images.

Is DeepSeek back on top?

❤️ — Yes, top-tier model!
🔥 — No, they are barely catching up