---
id: "deepseek/deepseek-r1"
also: ["deepseek/deepseek-r1", "deepseek-ai/DeepSeek-R1", "hf/deepseek-ai/DeepSeek-R1", "requesty/deepseek/deepseek-r1", "vercel/deepseek/deepseek-r1"]
cheapest_list_usd_per_m: "0.5 in, 2.15 out, at deepinfra on vercel"
cheapest_effective_usd_per_m: "0.4273 in, 2.15 out, at deepinfra on vercel"
context: 64000
max_output: 16000
input: ["text"]
output: ["text"]
released: "2025-01-20"
params: ["frequency_penalty", "include_reasoning", "max_completion_tokens", "max_tokens", "presence_penalty", "reasoning", "repetition_penalty", "response_format", "seed", "stop", "temperature", "tool_choice", "tools", "top_k", "top_p"]
endpoints: 7
gateways: "openrouter (2), requesty (2), vercel (2), hf (1)"
no_training: 5
trains: 0
retains_prompts: 0
intelligence: 11.4000
coding: 24.6000
agentic: 1.1000
---
# DeepSeek: R1

DeepSeek-R1 is an advanced, open-weights reasoning model by DeepSeek. Designed to compete with flagship reasoning models like OpenAI's o1, it focuses heavily on complex problem-solving in mathematics, coding, and science by "thinking" before it speaks, natively outputting a detailed, step-by-step chain of thought.