Skip to content
aiqbench

Google · model passport

Google: Gemma 3 12B

gemma-3-12b-it
  • Open weights
  • Gemma licence
  • 12.2B params
  • 131,072 context
  • Released 2025-03-12(curated)
9/27axes covered
Overall · LMArena
1,342#214 of 396
Best percentile
94.3HHEM Hallucination (Vectara)
Cohort standing
44.0median 45.8

What the data says

Benchmarked on 9 of 27 tracked axes. Every figure below is the model's own published result on that board.

Strongest axes

  • 94.3 pct — HHEM Hallucination (Vectara) (95.6%)

Uneven: 73.9 percentile points between its best and worst axis — 94.3 pct on HHEM Hallucination (Vectara) versus 20.4 pct on GPQA Diamond.

  • HHEM Hallucination (Vectara)94.3 pct95.6%Verify →
  • Overall46.2 pct1,342Verify →
  • Instruction following45.5 pct1,321Verify →
  • BFCL v4 (function calling)Dormant board42.7 pct30.4%Verify →
  • Hard prompts41.2 pct1,332Verify →
  • Coding29.9 pct1,316Verify →
  • LiveBench29.6 pct41.3%Verify →
  • AIME (math)28 pct16.7%Verify →
  • GPQA Diamond20.4 pct39.5%Verify →
  • WebDev ArenaNot benchmarked
  • Vision ArenaNot benchmarked
  • Factuality ArenaNot benchmarked
  • Search ArenaNot benchmarked
  • Document ArenaNot benchmarked
  • Agent ArenaNot benchmarked
  • FrontierMathNot benchmarked
  • SimpleQA VerifiedNot benchmarked
  • ARC-AGI-2Not benchmarked
  • Humanity's Last ExamNot benchmarked
  • SimpleBenchNot 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
  • 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

Repositorygoogle/gemma-3-12b-it
Parameters12.2B
ArchitectureGemma3ForConditionalGeneration
LicenceGemma
AccessGated — you must accept the repo's terms before downloading

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "google/gemma-3-12b-it"
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 IDgoogle/gemma-3-12b-it
Context length131,072 tokens
Max output16,384 tokens
Acceptstext, image
Input price$0.05 per 1M tokens
Output price$0.15 per 1M tokens

API parameters this endpoint supports

Frequency penaltyLogit biasMax tokensMin-pPresence penaltyRepetition penaltyResponse formatSeedStop sequencesStructured outputsTemperatureTool 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":"google/gemma-3-12b-it","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.

  • gemma-3-12b-it stands at 44 (raw mean 42) — measured on 9 of 27 axes.

    standingtemplateVerify →

  • gemma-3-12b-it stands at 44 versus a cohort median of 45.8 across 148 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 →