Gemma 4 31B vs Qwen3.5 27B
Wins 0 of 8 areas
—
Wins 5 of 8 areas
Coding · Agents · Images and charts · Math · Long documents
Qwen3.5 27B is the stronger all-rounder.Gemma 4 31B is cheaper.
Scores updated · 63 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 video14Qwen3.5 27B4 of 5 tests
- MathCompetition and research-level math03Qwen3.5 27B3 of 3 tests
- CodingWriting and fixing software23Qwen3.5 27B3 of 7 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own12Qwen3.5 27B2 of 3 tests
- Long documentsFinding answers in very long texts01Qwen3.5 27B1 of 1 test
- ReasoningHard problems that need careful thinking00Even0 each · 3 ties
- FactsGetting facts right instead of making them up11Even1 each · 1 tie
- Following instructionsDoing exactly what it is asked00Even0 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 4 31B costs 79% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3.5 27B pulls ahead
- Real work tasks from 44 professionsGDPVal+26.9points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+20.4points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+17.6points ahead
Where Gemma 4 31B pulls ahead
- Code for real scientific research problemsSciCode+6points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+3.8points ahead
- Finds the root cause of IT system incidentsAA IT-Bench SRE+1.8points ahead
Every test, side by side
All 63 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.5 27B
- SWE-bench VerifiedQwen3.5 27B by 20.45272.4+20.4
- SWE-bench MultilingualQwen3.5 27B by 17.651.769.3+17.6
- SWE-bench ProQwen3.5 27B by 15.535.751.2+15.5
- SciCodeGemma 4 31B by 645.539.5+6
- Terminal-Bench HardGemma 4 31B by 3.836.432.6+3.8
- LMArena · WebDevtie13651357tie
- LiveCodeBench v6tie8080.7tie
AgentsQwen3.5 27B
- GDPValQwen3.5 27B by 26.96.133+26.9
- MCP AtlasQwen3.5 27B by 11.257.268.4+11.2
- AA IT-Bench SREGemma 4 31B by 1.837.335.5+1.8
ReasoningEven
- CritPttie1.40.9tie
- Humanity's Last Examtie23.623.9tie
- GPQA Diamondtie85.785.8tie
FactsEven
- Vectara HHEM hallucination ratelower is betterGemma 4 31B by 4.77.412.1+4.7
- AA-Omniscience · Non-hallucinationQwen3.5 27B by 3.51518.5+3.5
- AA-Omniscience · Accuracytie2020.7tie
Images and chartsQwen3.5 27B
- CharXiv (RQ)Qwen3.5 27B by 11.667.979.5+11.6
- MathVistaQwen3.5 27B by 8.579.387.8+8.5
- LMArena · VisionGemma 4 31B by 37 rating points12771240+37 rating
- MMMUQwen3.5 27B by 1.980.482.3+1.9
- MMMU-ProQwen3.5 27B by 1.673.475+1.6
MathQwen3.5 27B
- IMOAnswerBenchQwen3.5 27B by 5.474.579.9+5.4
- HMMT Feb. 2026Qwen3.5 27B by 3.977.281.1+3.9
- AIME 2026Qwen3.5 27B by 2.589.291.7+2.5
Other results37 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.
- AA Agentic IndexQwen3.5 27B by 47.96.754.6+47.9
- τ²-Bench Telecom (AA run)Qwen3.5 27B by 3459.993.9+34
- WideSearchQwen3.5 27B by 25.935.261.1+25.9
- Claw Eval (pass@3)Qwen3.5 27B by 21.22546.2+21.2
- vectara_avg_summary_lengthQwen3.5 27B by 18.675.894.4+18.6
- MCPMarkQwen3.5 27B by 18.218.136.3+18.2
- Claw-Eval AvgQwen3.5 27B by 15.848.564.3+15.8
- QwenWebBenchGemma 4 31B by 129 rating points11971068+129 rating
- NL2RepoQwen3.5 27B by 11.815.527.3+11.8
- RealWorldQAQwen3.5 27B by 11.472.383.7+11.4
- QwenClawBenchQwen3.5 27B by 10.541.752.2+10.5
- Tool-DecathlonQwen3.5 27B by 10.321.231.5+10.3
- OmniDocBench 1.5Qwen3.5 27B by 8.880.188.9+8.8
- Artificial Analysis Coding IndexGemma 4 31B by 8.543.434.9+8.5
- AA IntelligenceQwen3.5 27B by 8.214.722.9+8.2
- C-EvalQwen3.5 27B by 7.982.690.5+7.9
- t2-benchGemma 4 31B by 7.486.479+7.4
- CC-OCRQwen3.5 27B by 5.375.781+5.3
- vectara_factual_consistencyGemma 4 31B by 4.792.687.9+4.7
- AA-OmniscienceQwen3.5 27B by 3.9-47.9-44+3.9
- AI2DQwen3.5 27B by 3.98992.9+3.9
- SkillsBench Avg5Qwen3.5 27B by 3.623.627.2+3.6
- HMMT Feb. 2025Qwen3.5 27B by 3.388.792+3.3
- SimpleVQAQwen3.5 27B by 3.152.956+3.1
- HallusionBenchQwen3.5 27B by 2.667.470+2.6
- MMMLUGemma 4 31B by 2.588.485.9+2.5
- HMMT Nov. 2025Qwen3.5 27B by 2.387.589.8+2.3
- MMBench EN-DEV-v1.1Qwen3.5 27B by 1.790.992.6+1.7
- DeepPlanningGemma 4 31B by 1.42422.6+1.4
- Terminal-Bench 2.0Gemma 4 31B by 1.342.941.6+1.3
- VITA-BenchGemma 4 31B by 1.14341.9+1.1
- MMLU-Protie85.286.1tie
- τ³-Benchtie67.568.4tie
- MMLU-Reduxtie93.793.2tie
- MATH-Visiontie85.686tie
- vectara_answer_ratetie10099.8tie
- SuperGPQAtie65.765.6tie
Questions people ask
Which is better, Gemma 4 31B or Qwen3.5 27B?
Qwen3.5 27B wins five of the eight areas we test: coding, agents, images and charts, math and long documents. Gemma 4 31B wins none, but costs 79% less. They are level on reasoning, facts and following instructions.
Which is better for coding?
Qwen3.5 27B. It wins 3 of the 7 coding tests both models report; Gemma 4 31B wins 2, and 2 are ties.
Which is cheaper?
Gemma 4 31B costs $0.17 per million input tokens and $0.40 per million output tokens; Qwen3.5 27B costs $0.30 and $2.40. That makes Gemma 4 31B about 79% cheaper for the same work.
How do you compare the two?
We use the 63 benchmark tests both models have published scores on. The verdict counts the 26 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 37 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.