Skip to content
AI Primer

DeepSeek-R1

Open-source reasoning LLM on par with OpenAI-o1.

DeepSeek-R1 is DeepSeek's first-generation open-source reasoning large language model release, a 671B-parameter MoE model trained with cold-start data and reinforcement learning to perform math, code, and reasoning tasks; DeepSeek states it can be accessed in the API by setting model=deepseek-reasoner.

Pricing

Official site · Aug 10, 2026, 1:10 PM
Input / 1M
$0.55
Output / 1M
$2.19
Cached input / 1M
$0.14

First-party DeepSeek-R1 API pricing page maps DeepSeek-R1 to model=deepseek-reasoner; input price recorded is cache-miss price, with cache-hit price recorded separately.

DeepSeek's official DeepSeek-R1 release/API page explicitly says to use DeepSeek-R1 by setting model=deepseek-reasoner and lists token API pricing: $0.14 per 1M input tokens for cache hits, $0.55 per 1M input tokens for cache misses, and $2.19 per 1M output tokens. Confidence is medium because the current general DeepSeek pricing page no longer lists DeepSeek-R1, but this first-party DeepSeek-R1 page remains public and explicit.

View source

Model Intelligence

Context window
128,000 tokens
Arena ranking
20
Benchmarkable
Yes
Model level
release
Intelligence Index
20.4
Math Index
76
MMLU Pro
0.85
GPQA
0.81
HLE
0.16
LiveCodeBench
0.77
SciCode
0.4
MATH-500
0.98
AIME
0.89
AIME 2025
0.76
IFBench
0.4
LCR
0.57
TerminalBench Hard
0.16
TAU2
0.37

Recent stories

1 linked story
AI PrimerAI Primer

Your daily guide to AI tools, workflows, and creative inspiration.

© 2026 AI Primer. All rights reserved.