'DeepSeek-V4-Flash-0731' has been released, an open model offering comparable performance to GPT-5.6 Luna at a lower price.

DeepSeek, a Chinese AI company, has released its AI model ' DeepSeek-V4-Flash-0731 ' as an open model under a license that allows commercial use.

deepseek-ai/DeepSeek-V4-Flash-0731 · Hugging Face

⚠️Note

🔷 DeepSeek-V4-Flash-0731 keeps the exact same model architecture and size as the preview version.
🔷 Today's upgrade applies ONLY to the DeepSeek-V4-Flash API. The DeepSeek-V4-Pro API and App/Web models remain unchanged for now.

The official release of DeepSeek-V4-Pro…

— DeepSeek (@deepseek_ai) July 31, 2026

DeepSeek-V4-Flash-0731 is an updated version of DeepSeek-V4-Flash , which was released in April 2026. It is a MoE model with a total of 284 billion parameters and 13 billion active parameters.

The following are the results of intelligence performance tests conducted by the third-party organization Artificial Analysis . DeepSeek-V4-Flash-0731 was rated as having performance equivalent to Google's Gemini 3.6 Flash, and was ranked as having performance third among open models, following Kimi K3 and GLM-5.2.

In terms of coding performance, it surpasses GLM-5.2.

In terms of agent performance, it significantly outperformed Gemini 3.6 Flash and beat OpenAI's GPT-5.6 Luna by 0.1 points.

DeepSeek-V4-Flash-0731 is characterized by its high performance relative to the model's size. The graph below shows the total number of model parameters on the horizontal axis and the intelligence performance on the vertical axis, demonstrating that DeepSeek-V4-Flash-0731 has high performance relative to the number of parameters.

The graph below shows the cost-effectiveness of the AI model, with the horizontal axis representing the cost per task and the vertical axis representing intelligence performance. DeepSeek-V4-Flash-0731 is even cheaper than GPT-5.6 Luna, which was reduced in price at the end of July 2026, demonstrating that it is a very cost-effective model.

DeepSeek-V4-Flash-0731 is freely available on Hugging Face and ModelScope, and is licensed under the MIT License. Hugging Face also offers numerous quantized versions created by volunteers.

Quantized Models for deepseek-ai/DeepSeek-V4-Flash-0731 – Hugging Face
https://huggingface.co/models?other=base_model:quantized:deepseek-ai/DeepSeek-V4-Flash-0731