Skip to content
TrustList
All launches
AI model launch

DeepSeek V4 Pro

DeepSeek's flagship leaves preview — 1M context, 384K output, built for tool-using agents.

Launched 2026-08-12AI model

About this launch

DeepSeek released the production version of its flagship, V4 Pro (build 0813), on 12 August 2026, ending a preview that ran nearly four months. It handles a 1M-token context and can emit up to 384,000 tokens in a single response, with a switchable thinking / non-thinking mode. The general-availability release is aimed squarely at agent work — tool use, code execution and multi-step workflows — scoring 87.9 on Terminal Bench 2.1, 62.7 on DeepSWE and 61.5 on NL2Repo. Buyers should note the pricing change that landed with it: from 16 August 2026 DeepSeek moved to peak/off-peak API rates, with peak-hour tokens costing double off-peak.

Discussion

Questions & answers about DeepSeek V4 Pro.

No questions yet. Be the first to ask about DeepSeek V4 Pro.