DeepSeek-R1 is the January 2025 MIT-licensed reasoning model with 671B total, 37B active parameters and 128K context. Its original API identity is historical.
DeepSeek-R1 is DeepSeek's January 2025 open reasoning model with 671B total, 37B active parameters, 128K context and MIT-licensed weights.
DeepSeek-R1 is DeepSeek’s open reasoning model released on January 20, 2025. The official checkpoint deepseek-ai/DeepSeek-R1 uses a mixture-of-experts architecture with 671 billion total parameters, 37 billion active parameters and a 128K context window.
DeepSeek used large-scale reinforcement learning in post-training to improve mathematics, code and natural-language reasoning. The company released the model weights and technical report under the MIT License and also published multiple smaller distill models based on R1-generated data.
At launch, DeepSeek exposed R1 through the deepseek-reasoner API alias. That alias was later upgraded to R1-0528, then to newer DeepSeek reasoning models, and the legacy alias itself was discontinued in July 2026. Therefore the January 2025 R1 checkpoint should not be assigned current DeepSeek API pricing.
R1 remains important for reasoning research, distillation and historical comparison. Self-hosting the full model is infrastructure intensive, so many developers will prefer smaller official distill models or current hosted models.
Solve this olympiad-style math problem and verify the result.Analyze this algorithm and derive its time complexity.Compare these two technical approaches and reason through the tradeoffs.Original DeepSeek-R1 is no longer the model behind DeepSeek's current API aliases. Historical R1 API pricing should be cleared; MIT weights remain downloadable.
Keep indexed as a major historical reasoning model, but separate the January 2025 R1 checkpoint from newer DeepSeek reasoning APIs and R1-0528.
NNNVIDIA Nemotron 3 Nano Omni 30B A3B Reasoning is an open-weight 31B multimodal MoE that understands text, images,…
Cohere Command A Reasoning is a 111B enterprise reasoning model with 256K context for agents, tool use, RAG…
Q8Qwen3.8-2.4T-A95B is Qwen’s open-weight flagship MoE with 2.4T total and 95B active parameters, a 1M-token context and hosted…
GOgpt-oss-20b is OpenAI’s open-weight 21B (3.6B active) reasoning model for fast local and private deployments, with a 131K…