Inference Index
←All models
Directory · Models · DeepSeek
DeepSeek R

DeepSeek R2

DeepSeek’s reasoning variant. Competes with o4 on math at a fraction of the cost.

Input / 1M
$0.55
Output / 1M
$2.19
Context
128K
Released
Jan 15, 2026

Benchmark profile

MMLU-ProHumanEvalMATHGPQAARC-AGIGSM8K
Profile
MMLU-Pro87.8%
HumanEval90.1%
MATH94.1%
GPQA71.2%
ARC-AGI94.1%
GSM8K94.1%

Scores

MMLU-Proknowledge
87.8%
HumanEvalcoding
90.1%
MATHmath
94.1%
GPQAreasoning
71.2%
ARC-AGIreasoning
94.1%
GSM8Kmath
94.1%
BIG-Bench Hardreasoning
71.2%
LiveBenchreasoning
87.8%

Editorial

Strengths
  • +Frontier math at open-weights pricing
  • +Long CoT traces
Trade-offs
  • −High output token usage inflates real cost

Price history (last 90 days)

Input / 1M
Output / 1M
04-1804-2705-0605-1605-2506-0306-13$0.00$0.73$1.46$2.19

What this costs to use

All task cards →
Compare across every model →

Recommended alternatives

Cheaper
GPT-5 Mini
OpenAI
$0.50/M · $2.00/M−8.8%
Stronger
DeepSeek V4
DeepSeek
$0.14/M · $0.28/M−85%
Direct competitor
Nova Pro
Amazon
$0.80/M · $3.20/M+46%

Where to buy

All providers →
Input pricing · $/1M tokens · sorted cheapest first