DeepSeek Models

The first and smallest model in DeepSeek's new architecture family — natively multimodal, and now what the deepseek-flash API name serves

DeepSeek's frontier MoE flagship, generally available since August 2026 with substantially stronger agentic tool use and code execution

DeepSeek's smaller, fast variant of V4 — same architecture at a fraction of the cost and latency

DeepSeek's latest flagship model matching GPT-5 performance with integrated tool-use thinking

DeepSeek's competition-focused variant (EXPIRED Dec 15, 2025 - was temporary API-only release)

DeepSeek's hybrid model combining V3 and R1 strengths

DeepSeek's upgraded reasoning model with 87.5% AIME accuracy and significantly reduced hallucinations

DeepSeek's updated V3 model with improved post-training using R1 techniques

DeepSeek's reasoning model competing with OpenAI o1, released under MIT license

DeepSeek's latest flagship model with enhanced capabilities and efficiency

DeepSeek's specialized coding model with advanced programming capabilities

DeepSeek's flagship MoE model with exceptional efficiency

DeepSeek's mathematics-specialized model