Gemma 3 27B vs Gemma 3 4B
Wins 5 of 7 areas
Coding · Reasoning · Facts · Images and charts · Following instructions
Wins 0 of 7 areas
—
Gemma 3 27B 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.
- CodingWriting and fixing software30Gemma 3 27B3 of 3 tests
- Images and chartsUnderstanding pictures, charts and video20Gemma 3 27B2 of 2 tests
- ReasoningHard problems that need careful thinking10Gemma 3 27B1 of 3 tests · 2 ties
- FactsGetting facts right instead of making them up21Gemma 3 27B2 of 3 tests
- Following instructionsDoing exactly what it is asked10Gemma 3 27B1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 2 ties
- Long documentsFinding answers in very long texts00Even0 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 37% 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.
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 27B
- SciCodeGemma 3 27B by 1623.37.3+16
- LiveCodeBench v6Gemma 3 27B by 10.229.118.9+10.2
- Terminal-Bench HardGemma 3 27B by 33.80.8+3
ReasoningGemma 3 27B
- GPQA DiamondGemma 3 27B by 13.742.829.1+13.7
- Humanity's Last Examtie4.45.3tie
- CritPttie00tie
FactsGemma 3 27B
- AA-Omniscience · Non-hallucinationGemma 3 27B by 6.17.91.8+6.1
- AA-Omniscience · AccuracyGemma 3 27B by 5.312.97.7+5.3
- Vectara HHEM hallucination ratelower is betterGemma 3 4B by 17.46.4+1
Images and chartsGemma 3 27B
Following instructionsGemma 3 27B
- IFBenchGemma 3 27B by 3.531.828.3+3.5
Other results25 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 27B by 31.598.867.3+31.5
- MMLU-ProGemma 3 27B by 2467.643.6+24
- Global-MMLU-LiteGemma 3 27B by 20.675.154.5+20.6
- InfoVQAGemma 3 27B by 20.670.650+20.6
- MMMLUGemma 3 27B by 20.670.750.1+20.6
- vectara_avg_summary_lengthGemma 3 27B by 1996.477.4+19
- MMLUGemma 3 27B by 18.576.958.4+18.5
- LiveCodeBenchGemma 3 27B by 17.129.712.6+17.1
- HumanEvalGemma 3 27B by 16.587.871.3+16.5
- MMMU (val) (Pass@1)Gemma 3 27B by 16.164.948.8+16.1
- AA-OmniscienceGemma 3 27B by 15.7-67.2-82.9+15.7
- BBHGemma 3 27B by 15.487.672.2+15.4
- MBPPGemma 3 27B by 11.274.463.2+11.2
- DocVQAGemma 3 27B by 10.886.675.8+10.8
- AI2DGemma 3 27B by 9.784.574.8+9.7
- ChartQAGemma 3 27B by 9.27868.8+9.2
- BIG-Bench Extra HardGemma 3 27B by 8.319.311+8.3
- Artificial Analysis Coding IndexGemma 3 27B by 7.410.12.7+7.4
- TextVQAGemma 3 27B by 7.365.157.8+7.3
- GSM8KGemma 3 27B by 6.795.989.2+6.7
- SimpleQAGemma 3 27B by 6104+6
- τ²-Bench Telecom (AA run)Gemma 3 27B by 5.510.55+5.5
- vectara_factual_consistencyGemma 3 4B by 192.693.6+1
- IFEvaltie90.490.2tie
- AA Intelligencetie4.94.8tie
Questions people ask
Which is better, Gemma 3 27B or Gemma 3 4B?
Gemma 3 27B wins five of the seven areas where both have results: coding, reasoning, facts, images and charts and following instructions. Gemma 3 4B wins none, but costs 37% less. They are level on agents and long documents.
Which is better for coding?
Gemma 3 27B. It wins 3 of the 3 coding tests both models report; Gemma 3 4B wins none.
Which is cheaper?
Gemma 3 27B costs $0.08 per million input tokens and $0.16 per million output tokens; Gemma 3 4B costs $0.05 and $0.10. That makes Gemma 3 4B about 37% 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 15 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 25 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.