- Published on
DeepSeek V4 Pro 0813 (on OpenRouter)
- Authors
- Name
- DeepHorus
DeepSeek has quietly released its newest AI model, DeepSeek V4 Pro 0813, available exclusively through API on OpenRouter as of August 12, 2026. The company has not posted an official announcement, leaving developers to discover the update via third-party platforms. This release follows the pattern of earlier DeepSeek models but raises questions about the company's communication strategy.
DeepSeek's previous models, deepseek-ai/DeepSeek-V4-Pro from April and deepseek-ai/DeepSeek-V4-Flash-0731 from July, both had open-weights releases. While the company has not confirmed whether the 0813 version will follow suit, the historical pattern suggests open weights are likely. The model is currently API-only, which may disappoint researchers hoping for local deployment.
According to community reports, benchmark scores for the new model were first distributed through DeepSeek's official WeChat group, then reposted to Reddit where moderators deleted the thread as low-effort, and eventually circulated as an ASCII-art table on Hacker News. One user testing the model observed notably different outputs for generated pelicans across low, medium, and high reasoning settings, a variation they had not seen from other models. These details point to the model's enhanced reasoning adaptability, though official benchmark documentation remains elusive.
The absence of a formal announcement page suggests DeepSeek may be prioritizing rapid iteration over conventional marketing. The API-only availability places DeepSeek in competition with other closed-API providers, while the likely open-weight release could strengthen its position among open-source advocates. The model's variable reasoning behavior may offer developers more flexible control but also adds complexity for benchmarking.
Community reaction has been a mix of enthusiasm and frustration. Enthusiasts praise the model's unique reasoning-level differences, while others criticize the fragmented distribution of benchmark data and the lack of official communication. The deleted Reddit post and reliance on Hacker News tables underline the community's resourcefulness in aggregating information.
DeepSeek has not set a timeline for open-weight release or a formal announcement. If the pattern continues, the next iteration may arrive within months, likely with a similar quiet rollout. Observers expect the company to eventually clarify its stance on open weights and provide proper documentation.
