Skip to content
aiqbench

DeepSeek · model passport

DeepSeek V3 2 Thinking

deepseek-v3-2-thinking
  • Open weights
  • MIT licence
  • Released 2026-10-13(first seen on LMArena)
7/27axes covered
Overall · LMArena
1,422#115 of 396
Best percentile
75.6LiveBench
Cohort standing
58.8median 60.8

What the data says

Benchmarked on 7 of 27 tracked axes. Every figure below is the model's own published result on that board.

Uneven: 46.8 percentile points between its best and worst axis — 75.6 pct on LiveBench versus 28.8 pct on WebDev Arena.

  • LiveBench75.6 pct66.2%Verify →
  • Instruction following75 pct1,419Verify →
  • Coding73.7 pct1,475Verify →
  • Hard prompts72.5 pct1,446Verify →
  • Overall71.2 pct1,422Verify →
  • Factuality Arena41.3 pct1,432Verify →
  • WebDev Arena28.8 pct1,362Verify →
  • Vision ArenaNot benchmarked
  • Search ArenaNot benchmarked
  • Document ArenaNot benchmarked
  • Agent ArenaNot benchmarked
  • GPQA DiamondNot benchmarked
  • AIME (math)Not benchmarked
  • FrontierMathNot benchmarked
  • SimpleQA VerifiedNot benchmarked
  • ARC-AGI-2Not benchmarked
  • Humanity's Last ExamNot benchmarked
  • SimpleBenchNot benchmarked
  • BFCL v4 (function calling)Not benchmarked
  • SWE-bench Verified (live board)Not benchmarked
  • Terminal-Bench 4.0Not benchmarked
  • MMLU-Pro (Artificial Analysis)Not benchmarked
  • LiveCodeBench (Artificial Analysis)Not benchmarked
  • IFBench (Artificial Analysis)Not benchmarked
  • HHEM Hallucination (Vectara)Not 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.

  • deepseek-v3.2-thinking stands at 58.8 (raw mean 62.6) — measured on 7 of 27 axes.

    standingtemplateVerify →

  • deepseek-v3.2-thinking stands at 58.8 versus a cohort median of 60.8 across 168 models released within 180 days — below 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 →