Skip to content
aiqbench

DeepSeek · model passport

DeepSeek: R1

deepseek-r1
  • Open weights
  • MIT licence
  • 684.5B params
  • 64,000 context
  • Released 2025-01-20(curated)
13/27axes covered
Overall · LMArena
1,398#152 of 396
Best percentile
89.1MMLU-Pro (Artificial Analysis)
Cohort standing
53.1median 42.5

What the data says

Benchmarked on 13 of 27 tracked axes. Every figure below is the model's own published result on that board.

Strongest axes

  • 89.1 pct — MMLU-Pro (Artificial Analysis) (84.4%)
  • 86.3 pct — LiveBench (71.6%)

Uneven: 65.5 percentile points between its best and worst axis — 89.1 pct on MMLU-Pro (Artificial Analysis) versus 23.6 pct on ARC-AGI-2.

  • MMLU-Pro (Artificial Analysis)89.1 pct84.4%Verify →
  • LiveBench86.3 pct71.6%Verify →
  • Instruction following65.4 pct1,397Verify →
  • LiveCodeBench (Artificial Analysis)65.3 pct61.7%Verify →
  • Hard prompts62.4 pct1,419Verify →
  • Overall61.9 pct1,398Verify →
  • Coding61.9 pct1,444Verify →
  • GPQA Diamond50.7 pct69.2%Verify →
  • AIME (math)39.5 pct53.3%Verify →
  • IFBench (Artificial Analysis)37.7 pct39.0%Verify →
  • HHEM Hallucination (Vectara)31.4 pct88.7%Verify →
  • SimpleBench24.1 pct30.9%Verify →
  • ARC-AGI-223.6 pct1.3%Verify →
  • WebDev ArenaNot benchmarked
  • Vision ArenaNot benchmarked
  • Factuality ArenaNot benchmarked
  • Search ArenaNot benchmarked
  • Document ArenaNot benchmarked
  • Agent ArenaNot benchmarked
  • FrontierMathNot benchmarked
  • SimpleQA VerifiedNot benchmarked
  • Humanity's Last ExamNot benchmarked
  • BFCL v4 (function calling)Not benchmarked
  • SWE-bench Verified (live board)Not benchmarked
  • Terminal-Bench 4.0Not benchmarked
  • SWE-bench VerifiedNot benchmarked
  • Terminal-BenchNot benchmarked

Percentiles are the model's standing within each board's own population. Axes marked "not benchmarked" have no published result for this model — that is an absence of data, not a low score.

How to run it

Run it yourself

Repositorydeepseek-ai/DeepSeek-R1
Parameters684.5B
ArchitectureDeepseekV3ForCausalLM
LicenceMIT

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "deepseek-ai/DeepSeek-R1"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

Ollama

We don't sync Ollama tags, so we won't print a command we can't verify — a wrong one would pull the wrong model. Search the Ollama library →

Unsloth fine-tuning recipes are model-specific and we don't derive them here. Unsloth docs →

Use it hosted

API model IDdeepseek/deepseek-r1
Context length64,000 tokens
Max output16,000 tokens
Input price$0.70 per 1M tokens
Output price$2.50 per 1M tokens

API parameters this endpoint supports

Frequency penaltyInclude reasoningMax tokensPresence penaltyReasoningRepetition penaltyResponse formatSeedStop sequencesTemperatureTool choiceTool callingTop-kTop-p

Example request

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek/deepseek-r1","messages":[{"role":"user","content":"Hello"}]}'

View on OpenRouter →

Identifiers, limits and prices are synced from the upstream source shown above. Any field we haven't synced is left out entirely rather than guessed.

  • deepseek-r1 stands at 53.1 (raw mean 53.8) — measured on 13 of 27 axes.

    standingtemplateVerify →

  • deepseek-r1 stands at 53.1 versus a cohort median of 42.5 across 141 models released within 180 days — above the peer median.

    cohorttemplateVerify →

Every claim is a computed fact from the current board data, phrased by a low-temperature model call and rejected by a validation gate if any number is not in the source row. How the leaderboards work →