Overview
- DeepSeek released a V4 preview in Pro and Flash versions as open source, with the rollout coinciding with OpenAI’s GPT-5.5 news.
- The new model introduces an up to one-million-token context window, a revamped architecture, and three reasoning modes that aim to handle long codebases and tougher math.
- DeepSeek V4-Pro lists pricing at $1.74 per million input tokens and $3.48 per million output tokens, while V4-Flash is $0.14 and $0.28, undercutting major U.S. rivals on cost.
- Heise reports experts place V4 roughly three to six months behind the very top models on overall ability, even as DeepSeek highlights strong coding results and claims high world-knowledge scores.
- Bloomberg notes a compute shortage is constraining Pro capacity, and DeepSeek plans new Huawei Ascend 950 clusters in the second half of the year that it says could ease supply and lower prices.