DeepSeek-V4-Pro-0813
DeepSeek · open weight · deepseek-v4-pro-0813
DeepSeek-V4-Pro-0813 is the GA / production release of DeepSeek-V4-Pro, with major upgrades for agentic coding and tool-use workloads. It is served behind the official API id `deepseek-v4-pro` and is distinct from catalog `deepseek-v4-pro-max`, which is the maximum reasoning-effort mode of Pro rather than this dated GA build. Same Pro MoE scale as peers (1.6T total / 49B activated). It supports low, high, and max reasoning effort, a 1M-token context window, up to 384K output tokens, and DeepSeek's OpenAI-compatible Chat Completions and Responses APIs. GA open weights are on Hugging Face at `deepseek-ai/DeepSeek-V4-Pro-0813` (MIT), distinct from the April preview repo `deepseek-ai/DeepSeek-V4-Pro`. New time-of-day pricing takes effect August 16: peak rates are $0.044 cached input, $1.32 uncached input, and $3.96 output per million tokens; off-peak rates are half those amounts.
Benchmark scores
| Benchmark | Score |
|---|---|
| Terminal-Bench 2.1 | 0.88 |
| CyberGym | 0.83 |
| Toolathlon | 0.74 |
| DSBench-FullStack | 0.71 |
| Humanity's Last Exam | 0.60 |
Pricing
- DeepInfra$1.30 / $2.60
- Together$1.32 / $3.96
- DeepSeek$1.32 / $3.96
- Novita$1.32 / $3.96
Input / output per 1M tokens
AA metrics
No Artificial Analysis link yet.
Arena Elo
- No Arena snapshot linked.