DeepYard
D

DeepSeek R1

11

Open-source reasoning model rivaling o1 at a fraction of the cost

Open Sourceopen-source

About

DeepSeek R1 is a powerful open-source reasoning model that achieves competitive performance with OpenAI's o1 series. The latest version (R1 0528) features 163K context and is available for free self-hosting or via API at just $0.40/1M input tokens. DeepSeek also offers distilled variants (Qwen 32B, Llama 70B) for efficient deployment. A breakthrough in open-source AI reasoning capabilities.

Details

Modalitytext
Context Window163.8K tokens
Parameters671B (MoE)
Release DateJan 20, 2025
API AvailableYes
Hostingself-hosted, api, together-ai, replicate
Output Speed90 tokens/sec
Time to First Token2000ms
Quality Index84/100
Coding Index72/100
Reasoning Index95/100

Benchmarks

Chatbot Arena ELO
1440/1500
MMLU-Pro
82/100
SWE-bench Verified
49.2/100
MATH-500
97.3/100
GPQA Diamond
71/100
HumanEval
91.6/100
AIME 2025
87.5/100

Tags

reasoningopen-sourcechain-of-thoughtself-hostedfine-tunablecost-effectiveopen-weight