DeepSeek · 2026-08-12 · major
DeepSeek V4 Pro 0813 — the 1.6T flagship leaves preview
DeepSeek V4 Pro 0813 is the general-availability build of DeepSeek's flagship mixture-of-experts model, ending a preview that began in April. Pricing holds at $0.435 in and $0.87 out per million tokens.
DeepSeek's 1.6-trillion-parameter flagship becomes a stable GA build after nearly four months in preview.
Quick facts
| Maker | DeepSeek |
|---|---|
| Model version | DeepSeek-V4-Pro-0813 |
| Parameters | 1.6T total, 49B active per token |
| Context window | 1M tokens |
| Price (input) | $0.435 / 1M tokens on a cache miss |
| Price (output) | $0.87 / 1M tokens |
| Availability | DeepSeek API and OpenRouter |
Pricing
| Input · cache miss | $0.435 / 1M tokens |
|---|---|
| Input · cache hit | $0.003625 / 1M tokens |
| Output | $0.87 / 1M tokens |
What is it?
DeepSeek V4 Pro 0813 is the general-availability version of DeepSeek V4 Pro, which had run as a preview since late April 2026. The API name does not change — calls to deepseek-v4-pro now route to the 0813 build. DeepSeek shipped it without a blog post, so the change shows up only in the API docs and on model pages such as OpenRouter.
How does it work?
The model is a mixture-of-experts network with 1.6 trillion total parameters and 49 billion active per token. Its hybrid attention design mixes a compressed sparse path with a heavily compressed one, which is what keeps a one-million-token context practical. Caching does the rest of the cost work: a cache hit costs $0.003625 per million input tokens, roughly 120x less than a cache miss.
Why does it matter?
Teams pinned to the preview build now get a stable flagship at unchanged rates. But the same pricing page carries a warning: DeepSeek plans to raise API prices 'in the near future, with a significant increase expected'. Anyone whose cost model rests on $0.87 per million output tokens should treat today's rate as temporary.
Who is it for?
teams running high-volume coding and long-context workloads
Frequently asked questions
- Is DeepSeek V4 Pro about to get more expensive?
- DeepSeek says on its own pricing page that it plans to raise overall API prices 'in the near future, with a significant increase expected', and advises users to plan usage accordingly. The current DeepSeek V4 Pro rates of $0.435 per million input tokens and $0.87 per million output tokens carried over unchanged from the preview period.
- Do I need to change my code for DeepSeek V4 Pro 0813?
- No code change is needed. DeepSeek routes the existing deepseek-v4-pro model name to the DeepSeek-V4-Pro-0813 build automatically, and the calling method stays the same. The DeepSeek API remains reachable through both the OpenAI ChatCompletions interface and the Anthropic interface.
- How is DeepSeek V4 Pro different from DeepSeek V4-Flash?
- DeepSeek V4 Pro is the large flagship at 1.6 trillion total parameters with 49 billion active per token, while V4-Flash is the smaller, cheaper 304B sibling that went to public beta on 31 July as DeepSeek-V4-Flash-0731. Flash scored 82.7 on Terminal Bench 2.1 in DeepSeek's own release note; Pro targets harder reasoning and long-context work.
- How long is the DeepSeek V4 Pro context window?
- DeepSeek V4 Pro handles one million tokens of context, which its model card attributes to a hybrid attention architecture combining compressed sparse attention with a heavily compressed path. That is enough to hold a large codebase or a long document set in a single request without external chunking.
Try it
Call model "deepseek-v4-pro" on the DeepSeek API