OpenRouter listed founder Liang Wenfeng's DeepSeek V4 Pro 0813 as a general-availability endpoint on August 12, with a 1-million-token context window and listed rates of $0.435 per million input tokens and $0.87 per million output tokens.
DeepSeek's July 31 change log said the V4 Pro API was unchanged and that its official release would follow soon. DeepSeek's homepage, captured the same day as OpenRouter's listing, likewise said V4-Pro had not changed.
The supplied materials establish a new versioned endpoint and API entry, but they do not establish an 0813-specific model, post-training, capability, latency, throughput or reliability change. DeepSeek has not explained the meaning of the 0813 suffix in the reviewed documentation.
Wenfeng founded Hangzhou-based DeepSeek in 2023 after co-founding and leading quantitative hedge fund High-Flyer, which has been associated with and funded the AI developer. He earned bachelor's and master's degrees in information and communication engineering from Zhejiang University, according to The Washington Post's profile. Public reporting has framed his motivation around original technical research and narrowing China's AI gap with US laboratories.
What developers can call
DeepSeek introduced the V4 family in preview on April 24 with Pro and Flash models designed around a 1-million-token context window. The April change log said developers could call V4 Pro through OpenAI ChatCompletions and Anthropic-compatible interfaces by setting the model parameter to deepseek-v4-pro
without changing the base URL.
OpenRouter now uses the versioned slug deepseek/deepseek-v4-pro-0813
. Its model page provides OpenAI-compatible chat completions and Responses endpoints as well as an Anthropic Messages endpoint. The listing exposes reasoning controls, streaming, tool calls and structured-output parameters.
DeepSeek documents a maximum output length of 384,000 tokens. That ceiling permits unusually long generations, though applications still have to manage the combined prompt, conversation history, tool results and output within the model's context allowance.
OpenRouter identifies DeepSeek as the endpoint's sole provider and says it forwards every request directly to DeepSeek. No alternate host is listed, so OpenRouter cannot shift an 0813 request to another provider if DeepSeek's endpoint is unavailable or degraded. That makes direct monitoring and fallback plans important for production deployments.
The release trail stops short of 0813
DeepSeek's July 31 change log documents how DeepSeek handled the smaller V4 Flash model. It says DeepSeek-V4-Flash-0731
retained the preview model's architecture and size while receiving additional post-training. The same notice explicitly limits that update to Flash and says the V4 Pro API, app and web models remained unchanged.
The reviewed change log contains no equivalent entry for V4 Pro 0813. OpenRouter calls 0813 the GA release, while DeepSeek's most recent notice in the supplied record said V4 Pro had not changed and its official release would follow. Those statements can coexist if OpenRouter versioned an existing deployment, if DeepSeek changed serving infrastructure without documenting it, or if DeepSeek had yet to publish a release note. The available evidence does not distinguish among those possibilities.
OpenRouter also displayed no latency or throughput measurements when the listing appeared. DeepSeek supplied no 0813-specific coding, reasoning, agent-task or reliability results in the reviewed materials. The GA label therefore describes availability on OpenRouter rather than verified performance relative to the previous V4 Pro endpoint.
Pricing favors controlled long-context tests
DeepSeek's pricing page lists current V4 Pro rates of $0.435 per million uncached input tokens, $0.003625 per million cache-read tokens and $0.87 per million output tokens. The page also warns of a future price increase, making current workload economics temporary.
The cache-read rate matters for applications that repeatedly submit the same repository, document collection, system prompt or conversation prefix. Operators still need to model cache-hit rates and output consumption because a long context allowance does not mean every request will cost the same. Agent loops can also multiply token use across planning, tool calls and retries.
OpenRouter's listing showed a usage-weighted effective input price below the posted rate on launch day because of caching. That figure reflects the traffic mix observed by OpenRouter and is not a guaranteed price for another customer's workload.
Early traffic spans chat and coding clients
The OpenRouter model page showed 394,000 tokens from SillyTavern, 255,000 from Velocity CLI, 57,000 from the pi coding agent, 33,000 from Janitor AI and 4,000 from Open WebUI on August 12. OpenRouter said there was not yet enough activity to display a traffic history.
Those counts show that chat interfaces and coding tools began testing the endpoint. They do not establish sustained adoption, customer retention or production reliability. OpenRouter separately says its broader marketplace serves more than 10 million users, processes over 200 trillion tokens a month and offers more than 400 models from over 70 providers. Those are OpenRouter-wide figures rather than usage figures for DeepSeek V4 Pro 0813.
The endpoint enters a market where other developers are also pitching long-context models for coding and agent work. MiniMax M3 supports a 1-million-token context window and adds native image, video and computer-use capabilities. OpenAI's GPT-5.3-Codex targets long-running coding and computer-use agents. Comparing them with 0813 requires task completion, latency, reliability and total workflow cost measurements that OpenRouter's launch-day listing does not provide.
For developers, the concrete event is access through a versioned OpenRouter slug backed solely by DeepSeek. The documented context window, API interfaces, output ceiling and current prices are sufficient for controlled tests. Teams considering a production migration still need to run their own workload evaluations because neither DeepSeek nor OpenRouter has tied the 0813 identifier to a measured model or serving change.