This is the general-availability build of V4 Pro, replacing the preview weights that had been quietly rotating under the same endpoint for weeks. The model is a 1.6T-parameter MoE with roughly 49B active parameters, a 1M-token context window, and up to 384K tokens of output. The gains are concentrated in agentic coding: DeepSeek reports DeepSWE climbing from 12.8 to 62.7, Cybergym from 52.7 to 83.3, and Terminal Bench 2.1 from 72.1 to 87.9 against the preview, and claims outright leads over Opus 4.8 on Terminal Bench 2.1, Cybergym, DeepSWE, and AutomationBench. Pricing stays aggressive — about $0.44 per million input tokens and $0.87 per million output, an order of magnitude below the frontier labs for a model DeepSeek positions within a few points of them. Worth knowing operationally: there's no new model ID, so requests already pointing at deepseek-v4-pro now resolve to the 0813 build with no change on your side. Fittingly for DeepSeek, the whole thing landed without a blog post or a tweet — the API documentation update was the announcement. A further price increase has been flagged for later.