Skip to content
aiqbench

DeepSeek-AI · model passport

DeepSeek V3

deepseek-v3
  • Open weights
  • DeepSeek licence
  • 684.5B params
  • Released 2024-12-26(curated)
13/27axes covered
Overall · LMArena
1,358#192 of 396
Best percentile
77.1HHEM Hallucination (Vectara)
Cohort standing
44.2median 42.0

What the data says

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

Weakest axes

  • 6.9 pct — SimpleBench (18.9%)

Uneven: 70.2 percentile points between its best and worst axis — 77.1 pct on HHEM Hallucination (Vectara) versus 6.9 pct on SimpleBench.

  • HHEM Hallucination (Vectara)77.1 pct93.9%Verify →
  • LiveBench63.6 pct60.5%Verify →
  • Overall51.8 pct1,358Verify →
  • MMLU-Pro (Artificial Analysis)51.7 pct75.2%Verify →
  • Instruction following50.8 pct1,343Verify →
  • Coding47.3 pct1,387Verify →
  • LiveCodeBench (Artificial Analysis)46.6 pct35.9%Verify →
  • Hard prompts45.5 pct1,350Verify →
  • GPQA Diamond40.3 pct56.5%Verify →
  • IFBench (Artificial Analysis)28.7 pct34.8%Verify →
  • AIME (math)26.5 pct15.8%Verify →
  • FrontierMath20.3 pct1.7%Verify →
  • SimpleBench6.9 pct18.9%Verify →
  • WebDev ArenaNot benchmarked
  • Vision ArenaNot benchmarked
  • Factuality ArenaNot benchmarked
  • Search ArenaNot benchmarked
  • Document ArenaNot benchmarked
  • Agent ArenaNot benchmarked
  • SimpleQA VerifiedNot benchmarked
  • ARC-AGI-2Not 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-V3
Parameters684.5B
ArchitectureDeepseekV3ForCausalLM
LicenceDeepSeek

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "deepseek-ai/DeepSeek-V3"
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 →

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-v3 stands at 44.2 (raw mean 42.8) — measured on 13 of 27 axes.

    standingtemplateVerify →

  • deepseek-v3 stands at 44.2 versus a cohort median of 42 across 138 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 →