Gemma 4 31B vs Qwen3.6 27B
Wins 2 of 8 areas
Reasoning · Following instructions
Wins 6 of 8 areas
Coding · Agents · Facts · Images and charts · Math · Long documents
Qwen3.6 27B is the stronger all-rounder.Gemma 4 31B is cheaper and better at reasoning.
Scores updated · 45 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- Images and chartsUnderstanding pictures, charts and video04Qwen3.6 27B4 of 4 tests
- CodingWriting and fixing software25Qwen3.6 27B5 of 7 tests
- MathCompetition and research-level math03Qwen3.6 27B3 of 3 tests
- AgentsCarrying out multi-step tasks on its own02Qwen3.6 27B2 of 3 tests · 1 tie
- FactsGetting facts right instead of making them up01Qwen3.6 27B1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts01Qwen3.6 27B1 of 1 test
- ReasoningHard problems that need careful thinking10Gemma 4 31B1 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked10Gemma 4 31B1 of 1 test
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 4 31B costs 87% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3.6 27B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+35.7points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+25.2points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+19.6points ahead
Where Gemma 4 31B pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+8points ahead
- Code for real scientific research problemsSciCode+2.7points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+1.6points ahead
Every test, side by side
All 45 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.6 27B
- SWE-bench VerifiedQwen3.6 27B by 25.25277.2+25.2
- SWE-bench MultilingualQwen3.6 27B by 19.651.771.3+19.6
- SWE-bench ProQwen3.6 27B by 17.835.753.5+17.8
- Terminal-Bench 2.1Qwen3.6 27B by 17.343.460.7+17.3
- LiveCodeBench v6Qwen3.6 27B by 3.98083.9+3.9
- SciCodeGemma 4 31B by 2.745.542.8+2.7
- Terminal-Bench HardGemma 4 31B by 1.636.434.8+1.6
AgentsQwen3.6 27B
- GDPValQwen3.6 27B by 18.36.124.4+18.3
- τ-Bench V3 · BankingQwen3.6 27B by 1.914.816.7+1.9
- Terminal-Bench 4.0tie00tie
ReasoningGemma 4 31B
- GPQA DiamondGemma 4 31B by 1.585.784.2+1.5
- Humanity's Last Examtie23.623.1tie
- CritPttie1.41.1tie
FactsQwen3.6 27B
- AA-Omniscience · Non-hallucinationQwen3.6 27B by 35.71550.7+35.7
- AA-Omniscience · Accuracytie2019.6tie
Images and chartsQwen3.6 27B
- CharXiv (RQ)Qwen3.6 27B by 10.567.978.4+10.5
- MathVistaQwen3.6 27B by 8.179.387.4+8.1
- MMMUQwen3.6 27B by 2.580.482.9+2.5
- MMMU-ProQwen3.6 27B by 1.273.474.6+1.2
MathQwen3.6 27B
- HMMT Feb. 2026Qwen3.6 27B by 7.177.284.3+7.1
- IMOAnswerBenchQwen3.6 27B by 6.374.580.8+6.3
- AIME 2026Qwen3.6 27B by 4.989.294.1+4.9
Following instructionsGemma 4 31B
- IFBenchGemma 4 31B by 875.667.6+8
Other results21 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.
- Claw Eval (pass@3)Qwen3.6 27B by 35.62560.6+35.6
- τ²-Bench Telecom (AA run)Qwen3.6 27B by 34.359.994.2+34.3
- QwenWebBenchQwen3.6 27B by 290 rating points11971487+290 rating
- AA-OmniscienceQwen3.6 27B by 27.9-47.9-20+27.9
- SkillsBench Avg5Qwen3.6 27B by 24.623.648.2+24.6
- Claw-Eval AvgQwen3.6 27B by 23.948.572.4+23.9
- NL2RepoQwen3.6 27B by 20.715.536.2+20.7
- Terminal-Bench 2.0Qwen3.6 27B by 16.442.959.3+16.4
- AA Agentic IndexQwen3.6 27B by 13.46.720.1+13.4
- RealWorldQAQwen3.6 27B by 11.872.384.1+11.8
- QwenClawBenchQwen3.6 27B by 11.741.753.4+11.7
- Artificial Analysis Coding IndexQwen3.6 27B by 10.343.453.7+10.3
- C-EvalQwen3.6 27B by 8.882.691.4+8.8
- AA IntelligenceQwen3.6 27B by 6.714.721.4+6.7
- CC-OCRQwen3.6 27B by 5.575.781.2+5.5
- HMMT Feb. 2025Qwen3.6 27B by 5.188.793.8+5.1
- HMMT Nov. 2025Qwen3.6 27B by 3.287.590.7+3.2
- SimpleVQAQwen3.6 27B by 3.252.956.1+3.2
- MMLU-ProQwen3.6 27B by 185.286.2+1
- SuperGPQAtie65.766tie
- MMLU-Reduxtie93.793.5tie
Questions people ask
Which is better, Gemma 4 31B or Qwen3.6 27B?
Qwen3.6 27B wins six of the eight areas we test: coding, agents, facts, images and charts, math and long documents. Gemma 4 31B wins reasoning and following instructions, and costs 87% less.
Which is better for coding?
Qwen3.6 27B. It wins 5 of the 7 coding tests both models report; Gemma 4 31B wins 2.
Which is cheaper?
Gemma 4 31B costs $0.17 per million input tokens and $0.40 per million output tokens; Qwen3.6 27B costs $0.60 and $3.60. That makes Gemma 4 31B about 87% cheaper for the same work.
How do you compare the two?
We use the 45 benchmark tests both models have published scores on. The verdict counts the 24 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 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.