DeepSeek Ships V4.1 Flash, a 552B Model for Coding

DeepSeek has a new model out, and it is a big one. V4.1 Flash is a 552B multimodal model built for coding and agentic workflows, and it takes both text and images.

The price is the real headline: $0.15 per 1M input tokens and $0.60 per 1M output tokens. On some benchmarks it scores 74.2, ahead of GPT-5.6 Sol and Claude Opus 5.

What Happened

DeepSeek released V4.1 Flash, a 552B model aimed at coding and agentic workflows. It processes text and images.

The numbers the post puts on it:

• 552B parameters.
• 74.2 on some benchmarks, beating GPT-5.6 Sol and Claude Opus 5.
• $0.15 per 1M input tokens and $0.60 per 1M output tokens.

It is up on Hugging Face at http://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash

Why It Matters

Token cost is what stops most people from leaving an agent loop running, and $0.60 per 1M output tokens is cheap enough that the bill stops being the reason you turn it off. If the benchmark numbers hold up, you get coding-grade output and image input in one model for less than you are probably paying now.

The Hugging Face page is the place to start if you want to see the model card and the weights yourself. Keep the word "some" in mind on that 74.2 — one score on one set of benchmarks is a starting point, not a verdict, so try it on the work you actually do before you build around it.