DeepSeek V4 Pro 0813
Points and comments are a snapshot, not live.
DeepSeek V4 Pro 0813 benchmarks show it competitive with Opus 4.8 but at 10-60x lower cost.
OpenRouter hosts DeepSeek V4 Pro 0813 with weighted average pricing of $0.03341/M input and $0.9011/M output tokens. Benchmarks include GPQA Diamond 92.8%, HLE 39.3%, and DeepSWE 62.7%. Top apps using the model include Hermes Agent, pi.dev, and Claude Code. Providers show uptime of 99.73% (3d average) and throughput up to 55 tokens per second. The model supports reasoning tokens, streaming, and tool calling via an OpenAI-compatible API.
What commenters are saying
Commenters largely agree V4 Pro outperforms DeepSeek Flash by about 5 percentage points on benchmarks, though Flash is much cheaper and faster. Some note Pro's superior reliability for code reviews, producing fewer mistakes and less output volume. Price comparisons show V4 Pro is 10-60x cheaper than Claude Opus 4.8 per task, especially with DeepSeek's aggressive caching. A few warn enterprise adoption of Chinese models may carry political risk, though providers in US/EU offer them as a service.