DeepSeek Ships V4 Pro, Claims Near-Parity With Claude at 1/46th the Price
What happened: Chinese AI firm DeepSeek released its V4 Pro (build 0813) model, claiming performance within 5% of Anthropic's Claude Fable 5 on its own nine-benchmark suite.
What happened: Chinese AI firm DeepSeek released its V4 Pro (build 0813) model, claiming performance within 5% of Anthropic's Claude Fable 5 on its own nine-benchmark suite. Pricing is $0.435 per million input tokens, compared to Fable 5's $10, making DeepSeek's model approximately 46 times cheaper. Independent benchmarks show V4 Pro competitive on some coding tasks, but trailing on advanced agentic tasks.
Why it matters: If DeepSeek's claims hold, V4 Pro could disrupt the economics of large language models, especially for high-volume users. However, all headline comparisons are vendor-reported, and independent labs have not yet replicated the results for the released build. The open-weight, MIT-licensed model could accelerate open-source AI adoption, but real-world parity with Claude remains unproven.
Source: Decrypt