DeepSeek Silently Releases V4-Pro-0813 with Output Price 57x Lower Than Claude Fable 5

My take: DeepSeek silently updated its flagship model on August 12: no blog post, no announcement, just a version change on its API pricing page. The price is $0.435 per million input tokens and $0.87 per million output tokens. Fable 5 costs $10 input and $50 output, making V4-Pro-0813 57x cheaper per output token and around 46x cheaper on blended rates.

DeepSeek reports a score of 87.9 on Terminal-Bench 2.1 against 88.0 for Fable 5, suggesting near-identical performance at a fraction of the cost. The required caveat: that benchmark was run by DeepSeek on its own model, with Fable 5 compared in a "fallback configuration" rather than its standard setup. When the company is both judge and party, we cannot know with certainty whether the figures are bias-free; it is worth waiting for independent evaluation before treating them as definitive. What a third party does measure: Artificial Analysis gives it 53 on its Intelligence Index, compared to 61 to 63 for models at the current frontier.

For teams processing large volumes of text that do not require absolute frontier performance, the pricing difference justifies a direct evaluation. The real gap between the best models keeps narrowing, and the cost of access keeps falling. Has your team compared what this month's bill would look like at your current model's rates versus a significantly cheaper alternative?