DeepSeek launches V4-Pro-0813 model at $0.87 per 1 million output tokens
This digest was compiled by AI from multiple sources — links to the originals are below.
DeepSeek has begun rolling out its V4-Pro-0813 model on its API and DeepSeek Chat, pricing it at $0.435 per 1 million input tokens and $0.87 per 1 million output tokens. The launch comes days after OpenAI cut GPT-5.6 Luna prices by up to 80 percent and hours after DeepSeek released its V4-Flash-0731 at $0.14 input and $0.28 output per million tokens. DeepSeek was second only to Anthropic in July token volume and faces a demand surge while running about 20,000 NVIDIA H100 GPUs.
V4-Pro-0813 Rollout
DeepSeek has begun distributing V4-Pro-0813 via its API and DeepSeek Chat. Input tokens are priced at $0.435 per 1 million, output tokens at $0.87 per 1 million. Preliminary results on WeChat show the model outcompetes Anthropic's Opus 4.8 on Terminal Bench 2.1, Cybergym, DeepSWE, and AutomationBench. The release is the company's latest attempt to challenge OpenAI and Anthropic.
Price War Escalation
OpenAI discounted GPT-5.6 Luna by up to 80 percent days earlier, cutting input to $0.20 and output to $1.20 per 1 million tokens. DeepSeek responded hours later with V4-Flash-0731, a 284-billion-parameter model priced at $0.14 per input and $0.28 per output million tokens. DeepSeek says that flash model achieves performance similar to Anthropic's Opus 4.8, which is widely believed to span multi-trillion parameters. The V4-Pro pricing now extends the lab's undercutting across another model tier.
Demand and Capacity
DeepSeek ranked second only to Anthropic in total token consumption in July and may reach the top spot in coming months. The lab is contending with a demand surge, with anecdotal reports of inference speeds slowing to a crawl. Its compute footprint is limited to roughly 20,000 NVIDIA H100 GPUs, constraining capacity. The token volume data and pricing moves put DeepSeek in direct competition with OpenAI and Anthropic.
What's Next
DeepSeek is expected to release more benchmark results as V4-Pro-0813 reaches full availability across its API and chat platforms. Whether the company can maintain response speeds and ascend to the top of monthly token volume with only about 20,000 NVIDIA H100 GPUs remains unclear.
4 sources
DeepSeek launches V4-Pro-0813 model at $0.87 per 1 million output tokens
