Inference
EN
DeepSeek

DeepSeek-R1

DeepSeek-R1 is a reasoning model from DeepSeek. 21 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2025-01
RankedEnough results for an index.
IndexAll categories, out of 100.38.9±5.0Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.18/sMedian 67/s
First tokenSeconds to the first output token.1.3 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.70Median $0.60
Output / 1MUS dollars per 1M output tokens.$2.50Median $2.55
ContextMaximum tokens in one request.64KMedian 262K
EloLMArena rating and rank.1373±5 #186Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
deepseek/deepseek-r1
Max outputMaximum output tokens per request.
16K
Knowledge cutoffLast date of training data.
31 Jul 2024
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
No
Cache read / 1MUS dollars per 1M cached tokens.
—
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Run locally

684.49B parameters
16-bit
1438.4 GB
8-bit
764.6 GB
4-bit
436.7 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.2 counted27.6204 / 218
CodingCode writing and repair.2 counted42.6174 / 209
ReasoningLogic problems and puzzles.4 counted39.8165 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.6 counted43.5177 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted28.556 / 63
MathMath problems.5 counted43.8115 / 168

Results

21 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Massive Multitask Language UnderstandingKnowledge90.8%2 / 34
MATH level 5Math93.1% ±0.713 / 88
MATH-500 Problem SetMath97.3%8 / 27
MATH 500Math92.2% ±2.418 / 59
Measuring Short-Form Factuality in Large Language ModelsKnowledge30.1%5 / 14
Massive Multitask Language Understanding ProfessionalKnowledge84.0%27 / 75
Artificial Analysis GPQA DiamondKnowledge81.3%94 / 199
Artificial Analysis Omniscience AccuracyKnowledge30.5%78 / 164
GPQA diamond · Epoch AIKnowledge71.7% ±3.1111 / 217
American Invitational Mathematics Examination 2024Math79.8%14 / 27
Critical Physics TasksReasoning1.4%113 / 214
MMLU ProKnowledge83.2% ±0.473 / 135
AIMEMath74.0% ±1.352 / 95
Artificial Analysis Humanity's Last ExamKnowledge15.8%123 / 222
Artificial Analysis MATH-500Math98.3%5 / 9
OTIS Mock AIME 2024-2025Math53.3% ±7.5118 / 200
Graduate-Level Google-Proof Q&AKnowledge71.5%87 / 141
LiveCodeBenchCoding70.2% ±1.188 / 142
Instruction-Following EvalInstruction83.3%45 / 70
Artificial Analysis Long Context ReasoningReasoning55.7%146 / 217
τ²-Bench Tool-Agent-User EvaluationAgentic36.5%107 / 149
Artificial Analysis IFBenchInstruction39.6%105 / 146
ARC-AGI-2 (semi-private)Reasoning1.3%103 / 131
Software Engineering Benchmark VerifiedCoding49.2%99 / 114
ARC-AGI-1 (semi-private)Reasoning15.8%111 / 126
Terminal-Bench 1.0Agentic13.8% ±3.945 / 47
Artificial Analysis MMLU-ProKnowledge84.9%15 / 15
MGSMMultilingual92.3% ±1.720 / 74
TaxEval v2Finance72.3% ±0.966 / 143
MedQAHealthcare90.8% ±0.544 / 95
CorpFin v2Finance54.1% ±1.097 / 131
ARC-AGI-2 (public eval)Reasoning0.3%99 / 122
LegalBenchLegal67.3% ±0.5133 / 149
Artificial Analysis Global-MMLU-LiteMultilingual86.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYVals AINo licence statedARC Prize FoundationTerms restrict reuse

History

5 releases
Index39.6 → 38.9
23 Oct 202612 Nov 2026
Output / 1M$2.50 → $2.50
23 Oct 202612 Nov 2026

Questions

How much does DeepSeek-R1 cost?
$0.70 per 1M input tokens, $2.50 per 1M output tokens.
How good is DeepSeek-R1?
Its index is 38.9 out of 100. That is place 215 of 283 ranked models.
What is DeepSeek-R1 best at?
Knowledge: 43.5 out of 100, place 177 of 268.
How fast is DeepSeek-R1?
18 output tokens per second, 1.3 s to the first token.
What is the context window of DeepSeek-R1?
64K tokens.
Is DeepSeek-R1 open source?
Yes. Its weights are published.
When was DeepSeek-R1 released?
January 2025.

Badge

DeepSeek-R1 on Inference360: place 215 of 283, index 38.9
HTML
<a href="https://inference360.ai/models/deepseek-r1"><img src="https://inference360.ai/badges/deepseek-r1.svg" alt="DeepSeek-R1 on Inference360: place 215 of 283, index 38.9" height="20"></a>
Markdown
[![DeepSeek-R1 on Inference360: place 215 of 283, index 38.9](https://inference360.ai/badges/deepseek-r1.svg)](https://inference360.ai/models/deepseek-r1)

Compare with

GPT-4.538.8Gemma 4 12B39.0Solar Pro 339.2Nemotron 3 Nano Omni 30B A3B38.6DeepSeek V3 032437.1

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from DeepSeek

All models
DeepSeek R1 Distill Qwen 32BUnrankedR1 Distill Llama 70BUnrankedDeepSeek V329.9DeepSeek V3 032437.1DeepSeek-R1-052844.6