Gemma 3 12B vs Gemma 3 4B
Wins 5 of 7 areas
Coding · Facts · Images and charts · Long documents · Following instructions
Wins 0 of 7 areas
—
Gemma 3 12B is the stronger all-rounder.Gemma 3 4B is cheaper.
Scores updated · 40 tests both models report · How we compare
Where each one wins
Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- FactsGetting facts right instead of making them up20Gemma 3 12B2 of 3 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video20Gemma 3 12B2 of 2 tests
- CodingWriting and fixing software10Gemma 3 12B1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts10Gemma 3 12B1 of 1 test
- Following instructionsDoing exactly what it is asked10Gemma 3 12B1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 2 ties
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Gemma 3 4B costs 25% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Gemma 3 12B pulls ahead
Where Gemma 3 4B pulls ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1.1points ahead
Every test, side by side
All 40 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemma 3 12B
- SciCodeGemma 3 12B by 9.116.47.3+9.1
- Terminal-Bench Hardtie0.80.8tie
ReasoningEven
- GPQA DiamondGemma 3 12B by 5.834.929.1+5.8
- Humanity's Last ExamGemma 3 4B by 1.14.25.3+1.1
- CritPttie00tie
FactsGemma 3 12B
- AA-Omniscience · AccuracyGemma 3 12B by 2.610.37.7+2.6
- Vectara HHEM hallucination ratelower is betterGemma 3 12B by 24.46.4+2
- AA-Omniscience · Non-hallucinationtie2.61.8tie
Images and chartsGemma 3 12B
Following instructionsGemma 3 12B
- IFBenchGemma 3 12B by 8.436.728.3+8.4
Other results26 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- vectara_answer_rateGemma 3 12B by 30.197.467.3+30.1
- MMLU-ProGemma 3 12B by 1760.643.6+17
- Global-MMLU-LiteGemma 3 12B by 1569.554.5+15
- InfoVQAGemma 3 12B by 14.964.950+14.9
- HumanEvalGemma 3 12B by 14.185.471.3+14.1
- WildBenchGemma 3 12B by 14.163.249.1+14.1
- BBHGemma 3 12B by 13.585.772.2+13.5
- vectara_avg_summary_lengthGemma 3 12B by 12.389.777.4+12.3
- LiveCodeBenchGemma 3 12B by 1224.612.6+12
- Arena HardGemma 3 12B by 11.843.631.8+11.8
- DocVQAGemma 3 12B by 11.387.175.8+11.3
- MMMU (val) (Pass@1)Gemma 3 12B by 10.859.648.8+10.8
- TextVQAGemma 3 12B by 9.967.757.8+9.9
- MBPPGemma 3 12B by 9.87363.2+9.8
- AI2DGemma 3 12B by 9.484.274.8+9.4
- ChartQAGemma 3 12B by 6.975.768.8+6.9
- AA-OmniscienceGemma 3 12B by 5.8-77.1-82.9+5.8
- τ²-Bench Telecom (AA run)Gemma 3 12B by 5.810.85+5.8
- BIG-Bench Extra HardGemma 3 12B by 5.316.311+5.3
- GSM8KGemma 3 12B by 5.294.489.2+5.2
- Artificial Analysis Coding IndexGemma 3 12B by 3.15.82.7+3.1
- SimpleQAGemma 3 12B by 2.36.34+2.3
- vectara_factual_consistencyGemma 3 12B by 295.693.6+2
- MM MTBenchGemma 3 12B by 1.56.75.2+1.5
- IFEvalGemma 3 4B by 1.388.990.2+1.3
- AA IntelligenceGemma 3 4B by 13.84.8+1
Questions people ask
Which is better, Gemma 3 12B or Gemma 3 4B?
Gemma 3 12B wins five of the seven areas where both have results: coding, facts, images and charts, long documents and following instructions. Gemma 3 4B wins none, but costs 25% less. They are level on agents and reasoning.
Which is better for coding?
Gemma 3 12B. It wins 1 of the 2 coding tests both models report; Gemma 3 4B wins none, and 1 is a tie.
Which is cheaper?
Gemma 3 12B costs $0.05 per million input tokens and $0.15 per million output tokens; Gemma 3 4B costs $0.05 and $0.10. That makes Gemma 3 4B about 25% cheaper for the same work.
How do you compare the two?
We use the 40 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 26 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.