DeepSeek model, API, reasoning, coding, and agent capability updates from the official docs.
DeepSeek model, API, reasoning, coding, and agent capability updates from the official docs.
DeepSeek API changelog spanning May 2024–August 2026. Key updates: DeepSeek-V4-Pro launch with enhanced agent capabilities (HLE 42.7, Terminal Bench 87.9), native Responses API, flexible thinking modes, and tiered API pricing. Sequential releases including V4-Flash, V3.2, V3.1, V3, and V2.5 show continuous improvements in reasoning and coding.
Source: official changelogDeepSeek API adds V4-Pro and V4-Flash through OpenAI ChatCompletions and Anthropic interfaces, with legacy names scheduled for retirement.
Source: official changelogDeepSeek upgrades deepseek-chat and deepseek-reasoner to V3.2 non-thinking and thinking modes, with a temporary V3.2-Speciale endpoint.
Source: official changelogDeepSeek upgrades both chat and reasoner model names to the V3.2-Exp experimental model track.
Source: official changelogDeepSeek V3.1-Terminus addresses reported issues, including language consistency and better Code Agent and Search Agent performance.
Source: official changelogDeepSeek V3.1 introduces a hybrid thinking and non-thinking architecture, improved reasoning efficiency, and stronger tool and agent task performance.
Source: official changelog