{"slug":"deepseek-r1","id":"deepseek-r1","type":"model","title":"DeepSeek R1","description":"Powerful open-source reasoning model that exceeds OpenAI o1 on AIME and MATH benchmarks. Transparent chain-of-thought reasoning at extremely low cost. MIT license. Updated with R1-0528 in May 2025.","last_updated":"2026-04-10","last_verified":null,"verification_status":"unverified","markdown_url":"/content/models/deepseek-r1.md","html_url":"/models/deepseek-r1","api_url":"/api/v1/models/deepseek-r1.json","content_hash":"82337e0cad76f5697b684824360eb063620b86b838f2550c40857a2257cc90f2","sha256":"82337e0cad76f5697b684824360eb063620b86b838f2550c40857a2257cc90f2","provider":"DeepSeek","pricing":{"input":"$0.55 / 1M tokens","output":"$2.19 / 1M tokens","note":"Also available open-source (MIT)"},"benchmarks":{"reasoning":92,"coding":88,"math":94,"writing":72,"multilingual":70,"speed":55},"tags":["deepseek","open-source","text"],"website":"https://deepseek.com","release_date":"2025-01","relationships":{"links":[],"related":[{"id":"deepseek-v3.2","title":"DeepSeek V3.2","type":"model","html_url":"/models/deepseek-v3.2","markdown_url":"/content/models/deepseek-v3.2.md","shared_tags":["deepseek","open-source","text"],"score":7},{"id":"cohere-tiny-aya","title":"Cohere Tiny Aya 3.35B","type":"model","html_url":"/models/cohere-tiny-aya","markdown_url":"/content/models/cohere-tiny-aya.md","shared_tags":["open-source","text"],"score":4},{"id":"command-r-plus","title":"Command R+","type":"model","html_url":"/models/command-r-plus","markdown_url":"/content/models/command-r-plus.md","shared_tags":["open-source","text"],"score":4},{"id":"provider-deepseek","title":"DeepSeek Provider Profile","type":"provider","html_url":"/providers/deepseek","markdown_url":"/content/providers/deepseek.md","shared_tags":["deepseek","open-source"],"score":4},{"id":"falcon-3","title":"Falcon 3","type":"model","html_url":"/models/falcon-3","markdown_url":"/content/models/falcon-3.md","shared_tags":["open-source","text"],"score":4},{"id":"gemma-3","title":"Gemma 3","type":"model","html_url":"/models/gemma-3","markdown_url":"/content/models/gemma-3.md","shared_tags":["open-source","text"],"score":4}],"explicit":{}},"metadata":{"title":"DeepSeek R1","type":"model","id":"deepseek-r1","provider":"DeepSeek","model_type":"open-source","release_date":"2025-01","description":"Powerful open-source reasoning model that exceeds OpenAI o1 on AIME and MATH benchmarks. Transparent chain-of-thought reasoning at extremely low cost. MIT license. Updated with R1-0528 in May 2025.","last_updated":"2026-04-10","context_window":"128K tokens","website":"https://deepseek.com","license":"MIT","modality":["text"],"tags":["deepseek","open-source","text"],"pricing":{"input":"$0.55 / 1M tokens","output":"$2.19 / 1M tokens","note":"Also available open-source (MIT)"},"benchmarks":{"reasoning":92,"coding":88,"math":94,"writing":72,"multilingual":70,"speed":55},"parameters":"671B total (37B active)","hardware_requirements":"8x A100 80GB (FP16); 2x A100 with Q4 quantization","best_for":["Mathematical reasoning","Code generation","Scientific analysis","Budget-conscious deployment"]},"content_text":"# DeepSeek R1\n\nThe open-source reasoning model that changed the game. DeepSeek R1 beat OpenAI's o1 on AIME and MATH benchmarks, scoring 94/100 in math -- and it does it under an MIT license at $0.55/$2.19 per million tokens. That's roughly 5% of what GPT-5.4 Thinking costs for math performance that's in the same conversation.\n\nR1's transparent chain-of-thought is both a feature and a constraint. You can see exactly how the model reasons through a problem, which is invaluable for education, research, and debugging. But the thinking process is slow (55/100 speed) and the outputs are less refined than what you get from proprietary models. The R1-0528 update improved stability, but this is still a model optimized for getting the right answer, not for presenting it beautifully.\n\nThe profile is extremely spiky. Math (94) and reasoning (92) are near-frontier. Writing (72) and multilingual (70) are genuinely weak. R1 will solve a differential equation better than most proprietary models, then produce a mediocre summary of its own solution. Self-hosting requires the same 8x A100 setup as V3.2, or you can use the API and let DeepSeek handle infrastructure.\n\n**When to pick something else:** For anything involving writing, conversation, or multilingual work, use literally any other model on this list. For the absolute ceiling on reasoning, GPT-5.4 Thinking (98/100) still leads, though at 20x the cost. For general-purpose coding and reasoning without the writing penalty, DeepSeek V3.2 is the more balanced sibling.","content_length":2472,"generated_at":"2026-04-24"}