DeepSeek V4 Pro 0813がOpenRouter経由で利用可能にDeepSeek V4 Pro 0813 (on OpenRouter)

DeepSeek has quietly made its latest large language model, DeepSeek V4 Pro 0813, available for use, but only through an API and initially without any official announcement of its own. The rollout is notable both for the model's lineage, since DeepSeek has become one of the most closely watched labs in the open-weight space, and for the unusual way it surfaced, with developers pointing to OpenRouter as the most reliable way to access it.

According to a post from Simon Willison, who tracks new model releases closely, the model is currently accessible via API only. He linked to OpenRouter rather than a first-party page because DeepSeek does not appear to have published an obvious announcement for the new model. That absence is worth flagging: for a lab whose earlier releases generated significant attention, the lack of a formal launch page makes it harder to confirm specifics such as context length, pricing tiers, and intended use cases directly from the source.

OpenRouter, for readers unfamiliar with it, is a routing and aggregation service that exposes many different models from many providers behind a single, unified API. That design lets developers switch between models, including Anthropic's Claude, OpenAI's GPT family, Google's Gemini, Meta's Llama variants, and numerous others, without rewriting integration code for each vendor. For a model that lacks a prominent official landing page, an aggregator like OpenRouter can effectively become the practical point of discovery and access.

One of the most significant open questions concerns open weights. DeepSeek built much of its reputation on releasing model weights that developers could download and run on their own hardware, a practice that distinguishes it from labs that keep their frontier models strictly behind APIs. Willison notes that he has not been able to confirm whether DeepSeek plans to release open weights for V4 Pro 0813. Until that is clarified, the model appears to be available only as a hosted service, which would be a departure from the pattern that made the company well known within the local-LLM community.

The "0813" suffix in the name appears to follow a date-based versioning convention, a labeling style several labs have adopted to distinguish incremental checkpoints of a model. This makes it likely that DeepSeek intends to iterate on the V4 Pro line over time, though the company has not detailed a roadmap. Without an official announcement, benchmark claims and architectural details remain unconfirmed, and readers should treat any performance comparisons circulating in the interim with caution.

DeepSeek rose to prominence with earlier releases in its V3 and R1 lines, which drew attention for competitive performance relative to their reported training costs and for the decision to make weights broadly available. Those releases fed into a wider industry conversation about whether open or openly licensed models could keep pace with proprietary systems from larger, better-funded competitors. An API-only V4 Pro release, if it stays that way, would run somewhat counter to that narrative, though it is also common for labs to offer hosted access first and publish weights later.

For developers, the immediate practical takeaway is that experimentation with V4 Pro 0813 is possible today through OpenRouter's unified interface, subject to whatever pricing and rate limits the routing service and DeepSeek set. Teams that specifically require self-hosting, on-premises deployment, or the ability to fine-tune on private data will need to wait for confirmation on open weights before committing. Those constraints matter for regulated industries and privacy-sensitive workloads where sending data to a third-party API is not an option.

As with any release that arrives without formal documentation, the situation may change quickly. DeepSeek could publish a full announcement, additional model variants, or open weights in the days ahead, or it could keep V4 Pro as a hosted offering. For now, the confirmed facts are narrow: the latest DeepSeek Pro model exists, it is reachable via API, OpenRouter is the practical access route, and the company's open-weight intentions remain unstated. Anyone evaluating the model for production use should watch for an official DeepSeek page that clarifies licensing, pricing, and deployment options before making longer-term architectural decisions.