DeepSeek has begun rolling out its V4-Pro-0813 model on its own programming interface and chat app, priced at $0.435 per million input tokens and $0.87 per million output tokens. Early benchmark figures circulating on WeChat, which put it ahead of Anthropic's Opus 4.8 on several coding and automation tests, have not been independently verified. The lab is reported to be straining under demand with roughly 20,000 Nvidia H100 chips, and users have described slow response times.
What changed
DeepSeek's top-tier option was its April V4 preview; rival OpenAI had just cut GPT-5.6 Luna output pricing to $1.20 per million tokens.
What it unlocks
Access to a high-end coding and agent model at well below the price of comparable Western top-tier models, via DeepSeek's API and chat app.
- $0.435 per 1M input tokens
- $0.87 per 1M output tokens
- 1M token context, 1.6T params
- +15.8% on Terminal Bench vs April
What you need to act on it
- DeepSeek API access or DeepSeek Chat
- tolerance for a China-hosted provider under load
Sources