DeepSeek

Post-trained 1017 release with stronger coding and agent results, faster task completion, shorter outputs, image input, and a 1M context window.

Post-trained 0731 release with major gains across coding, repository work, tool use, and full-stack tasks, plus a 1M context window.

Official 0813 Pro release with major gains across coding, repository work, tool use, and agent tasks, plus hybrid thinking and a 1M context window.

Flagship MoE LLM with 1.6T total / 49B active parameters and native 1M context for advanced math, logical inference, and specialized coding.

Autoregressive framework on the Janus Pro 7B model that unifies multimodal understanding and image generation in one architecture.