DeepSeek V4 Flash 0731 (Deep Infra)

On July 31, 2026, DeepSeek released DeepSeek-V4-Flash-0731 as an open model under a license that permits commercial use, with weights available on Hugging Face and ModelScope. The article states the model is a Mixture-of-Experts design with 284 billion total parameters and 13 billion active parameters, and DeepSeek's o According to third-party benchmarking by Artificial Analysis cited in the article, DeepSeek-V4-Flash-0731 was rated comparable in intelligence to Google's Gemini 3.6 Flash and ranked third among open models behind Kimi K3 and GLM-5.2. It reportedly surpasses GLM-5.2 in coding performance and edges past OpenAI's GPT-5.6

On July 31, 2026, DeepSeek released DeepSeek-V4-Flash-0731 and began a public beta of its API, with per-million-token pricing of $0.14 for cache-miss input and $0.28 for output. According to the article, those rates have been in place since before; what changed in the 0731 update is the model's capabilities and API fea The 0731 release is described as a re-post-training of the architecture inherited from the April 24 Preview: 280 billion total parameters with 13 billion active under a Mixture-of-Experts design, a 1-million-token context window, and a maximum output of 384,000 tokens. The update applies only to the V4 Flash API; the V

Better Stack's guide explicitly covers DeepSeek-V4-Flash-0731, confirming DeepSeek published the model on July 31, 2026 alongside a public beta of the official V4-Flash API. It details the architecture as a 284-billion-parameter Mixture-of-Experts model with 13 billion active parameters per token, a 1-million-token con The article reports launch pricing of $0.14 per million input tokens on cache miss, $0.0028 per million on cache hit, and $0.28 per million output tokens, while confirming the model weights are released under the MIT license on Hugging Face. It frames V4 Flash as occupying a cost-intelligence curve position below OpenA