Gemma 3 12B IT — Benchmarks

Benchmark scores for Gemma 3 12B IT aggregated from public leaderboards (8 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

74.2

90% range 68.6–79.8 · from 7 boards

CodingNot enough data yet
AgentsNot enough data yet
MathNot enough data yet
Science71.690% range 60–83
Reasoning66.490% range 52–81

Gemma 3 12B IT reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

GPQA Diamond
GPQA Diamond is a set of extremely hard, graduate-level science questions (physics, chemistry, biology) written by domain experts and filtered so that skilled non-experts with web access still fail. It measures genuine reasoning rather than memorization.
39.5%
39.5%
median 55.7%open leader 93.2%
Overall best: Gemini 4 Argon (high) 96.4%
27th percentile · #73 of 100 open · #266 of 309 all
#73 of 100
#266 of 309
Independent
Epoch AI · Mar 2025
CritPt
CritPt (Critical Points) tests graduate-level physics problem-solving on research-style questions spanning mechanics, electromagnetism, quantum mechanics, and more. It was created by physicists to probe whether a model can reason through multi-step physics rather than pattern-match textbook answers.
0.0%
0.0%
median 0.9%open leader 26.6%
Overall best: Gemini 4 Carbon (high) 36.0%
19th percentile · #56 of 84 open · #200 of 244 all
#56 of 84
#200 of 244
Lab-reported
Epoch AI · Mar 2025
DTBench
DTBench is a set of handcrafted, expert-validated multiple-choice questions on decision theory — Newcomb-like problems and related puzzles — built by the Conceptual Reasoning Index project with Anthropic. It tests whether a model can follow an abstract argument rather than recall a fact.
48.8%
48.8%
median 68.7%open leader 93.9%
Overall best: Claude Opus 5.5 Max 98.9%
9th percentile · #64 of 70 open · #223 of 235 all
#64 of 70
#223 of 235
Lab-reported
Epoch AI · Mar 2025
LMCA
LMCA (Language Model Conceptual Argumentation) scores models on expert-rated arguments across philosophy, decision theory, and AI risk — domains where there is no empirical feedback loop to check an answer against. It is part of the Conceptual Reasoning Index, built with Anthropic by researchers Emery Cooper and Caspar Oesterheld.
4.5%
4.5%
median 25.5%open leader 55.5%
Overall best: Claude Sonnet 5.5 Max 68.3%
2nd percentile · #55 of 56 open · #196 of 197 all
#55 of 56
#196 of 197
Lab-reported
Epoch AI · Mar 2025

Gemma 3 12B IT coding benchmarks

Writing, editing, and fixing real code — measured by whether the resulting program actually runs and passes tests.

SciCode
SciCode asks a model to write real scientific-computing code — implementing multi-step numerical methods drawn from physics, chemistry, biology, and math research — and checks the output against reference solutions. It was built by a team of scientists and ML researchers to test coding ability on genuine research problems, not typical software tasks.
16.4%
16.4%
median 40.1%open leader 60.9%
Overall best: Claude Opus 5.5 Max 66.9%
3rd percentile · #70 of 72 open · #215 of 217 all
#70 of 72
#215 of 217
Lab-reported
Epoch AI · Mar 2025

Gemma 3 12B IT math benchmarks

Symbolic and quantitative problem-solving, from competition math to multi-step calculation.

AIME 2024/2025
AIME (American Invitational Mathematics Examination) is a prestigious high-school competition of hard, integer-answer problems. It's a widely-cited yardstick for multi-step mathematical reasoning, here on the 2024–2025 papers.
16.7%
16.7%
median 54.4%open leader 99.4%
Overall best: Claude Fable 5 (high) 100.0%
36th percentile · #57 of 88 open · #236 of 292 all
#57 of 88
#236 of 292
Independent
Epoch AI · Mar 2025

Gemma 3 12B IT knowledge benchmarks

Breadth and depth of factual knowledge across many subjects, paired with the reasoning to apply it.

MMLU-Pro
MMLU-Pro is a harder, cleaned-up successor to MMLU with ten answer choices and more reasoning-heavy questions across 14 subjects, measuring broad knowledge and reasoning together.
60.6%
60.6%
median 52.0%open leader 88.0%
Overall best: Gemini-3.1-Pro 91.2%
61st percentile · #57 of 146 open · #136 of 257 all
#57 of 146
#136 of 257
Independent
TIGER-Lab

Gemma 3 12B IT instruction following benchmarks

How precisely a model obeys explicit, verifiable instructions — format, length, and content constraints — and, more broadly, how much people prefer its responses head-to-head in blind human voting.

LMArena Text
Arena (arena.ai, formerly LMArena / Chatbot Arena) ranks models by millions of blind, head-to-head human votes on real prompts. The Arena Score shown here is a Bradley-Terry rating computed with Arena's default style control, so higher means people preferred that model's answers more often — it's a relative rating, not a percentage, so compare it only to other models on this board.
1341.7
1341.7
median 1321.4open leader 1488.2
Overall best: Gemini 4 Carbon (High) 1527.5
56th percentile · #84 of 188 open · #227 of 408 all
#84 of 188
#227 of 408
Arena
Arena · Oct 2026

Score vs VRAM in the 24 GB tier

Gemma 3 12B IT
Other models
Best Score for its VRAM
x: VRAM (GB) · y: llmrun Score

#28 of 36 that fit 24 GB

Each dot is a model at the best quant that fits a 24 GB GPU. The line steps up to the best Score available at each amount of VRAM; a dot below it is beaten by a model that needs no more memory.

Full 24 GB ranking →

Gemma 3 12B IT vs its neighbours

Same boards for the nearest open models per llmrun Score and the best model that fits 24 GB. The highest score in each row is marked.

Board	Gemma 3 12B ITthis model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	Gemma 3 27B ITnearest by ScoreQ4_K_M · 22 GB	Mistral Small 3.1 24B Instruct 2503nearest by ScoreQ4_K_M · 18 GB	Gemma 2 27B ITnearest by ScoreQ4_K_M · 20 GB
llmrun Score	74.2	122.0	79.4	79.0	72.7
GPQA Diamond	39.5%	—	47.7%	47.5%	36.5%
CritPt	0.0%	5.4%	0.0%	0.0%	—
DTBench	48.8%	88.0%	52.5%	58.6%	—
LMCA	4.5%	41.4%	12.3%	—	—
SciCode	16.4%	46.6%	23.3%	27.8%	—
AIME 2024/2025	16.7%	—	22.5%	5.8%	1.4%
MMLU-Pro	60.6%	—	67.5%	—	56.5%
LMArena Text	1341.7	1435.6	1365.3	1302.6	1289.5
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 6 boards, Arena: 1 board, TIGER-Lab: 1 board) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.