DeepSeek Formally Releases V4-Pro-0813, Raises API Prices, Open-Sources Agent Harness

DeepSeek officially released its V4 Pro model — designated V4-Pro-0813 — on August 13, moving the model from preview (since April) to general availability across its API, app, and web channels, Reuters and Caixin reported. The GA release sharply improves agent-task benchmarks versus the April preview: Terminal Bench 2.1 rose from 72.1 to 87.9, DeepSWE from 12.8 to 62.7, and the Artificial Analysis Intelligence Index from 45 to 53, ranking behind Claude Opus 5, Kimi K3, and Qwen 3.8 Max but ahead of most Western mid-tier models. The model supports a 1-million-token context window and up to 384,000-token outputs in either thinking or non-thinking mode. DeepSeek simultaneously announced it will raise API pricing and introduce peak/off-peak billing effective 16:00 UTC August 16 — off-peak input rising from $0.435 to $0.66 per million tokens and output from $0.87 to $1.98, with peak rates double that — ending the aggressive promotional pricing that powered its 2025-2026 price war. The company also open-sourced "DeepSeek Harness v0.1," an MIT-licensed agent framework that drew interest from over 700 projects within three days. DeepSeek is separately reported to be raising a new funding round at a roughly $74 billion valuation, more than 10x its June 2025 round, and plans to roughly double headcount to compete with Moonshot AI, Zhipu AI, MiniMax, Alibaba, and ByteDance.