DeepSeek quietly released the general-availability version of its V4 Pro model (build 0813) on Wednesday, upgrading from a preview build that had been circulating since April with no official announcement. According to the company’s own benchmarks across ten agent benchmarks, Anthropic’s Claude Fable 5 outperforms V4 Pro by only 5.3% on average, with gaps of just 2.8% when excluding one outlier test. On pricing, V4 Pro costs $0.435 per million input tokens and $0.87 per million output tokens, compared to $10 and $50 for Fable 5, making it roughly 46 times cheaper. The cost per completed task gap widens to 105 times cheaper according to Artificial Analysis. The model weights are MIT-licensed and available on Hugging Face for independent verification.
Source: Read the original article
