Gemma 4 31B vs Qwen3.8 27B
Wins 0 of 7 areas
—
Wins 6 of 7 areas
Coding · Agents · Reasoning · Images and charts · Long documents · Following instructions
Qwen3.8 27B is the stronger all-rounder.Gemma 4 31B is cheaper.
Scores updated · 50 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 software07Qwen3.8 27B7 of 7 tests
- AgentsCarrying out multi-step tasks on its own05Qwen3.8 27B5 of 5 tests
- ReasoningHard problems that need careful thinking03Qwen3.8 27B3 of 3 tests
- Images and chartsUnderstanding pictures, charts and video02Qwen3.8 27B2 of 3 tests · 1 tie
- Long documentsFinding answers in very long texts01Qwen3.8 27B1 of 2 tests · 1 tie
- Following instructionsDoing exactly what it is asked01Qwen3.8 27B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Following instructions rests on a single test.
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 84% 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 Qwen3.8 27B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+54.7points ahead
- Real work tasks from 44 professionsGDPVal+40points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+36.4points ahead
Where Gemma 4 31B pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+4.4points ahead
Every test, side by side
All 50 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.8 27B
- Terminal-Bench 2.1Qwen3.8 27B by 36.443.479.8+36.4
- SWE-bench VerifiedQwen3.8 27B by 31.746.177.8+31.7
- SWE-bench MultilingualQwen3.8 27B by 30.84373.8+30.8
- SWE-bench ProQwen3.8 27B by 2635.761.7+26
- LMArena · WebDevQwen3.8 27B by 225 rating points13651590+225 rating
- LiveCodeBench v6Qwen3.8 27B by 10.38090.3+10.3
- SciCodeQwen3.8 27B by 1.145.546.6+1.1
AgentsQwen3.8 27B
- GDPValQwen3.8 27B by 40646+40
- τ-Bench V3 · BankingQwen3.8 27B by 33.214.848+33.2
- BrowseCompQwen3.8 27B by 24.63761.6+24.6
- MCP AtlasQwen3.8 27B by 23.154.277.3+23.1
- Terminal-Bench 4.0Qwen3.8 27B by 5.605.6+5.6
ReasoningQwen3.8 27B
- Humanity's Last ExamQwen3.8 27B by 10.323.633.9+10.3
- GPQA DiamondQwen3.8 27B by 4.885.790.5+4.8
- CritPtQwen3.8 27B by 41.45.4+4
FactsEven
- AA-Omniscience · Non-hallucinationQwen3.8 27B by 54.71569.7+54.7
- AA-Omniscience · AccuracyGemma 4 31B by 4.42015.6+4.4
Images and chartsQwen3.8 27B
- CharXiv (RQ)Qwen3.8 27B by 22.367.990.2+22.3
- MMMU-ProQwen3.8 27B by 2.973.476.3+2.9
- LMArena · Visiontie12791271tie
Long documentsQwen3.8 27B
- AA-LCRQwen3.8 27B by 12.369.782+12.3
- LongBench v2tie52.952.3tie
Following instructionsQwen3.8 27B
- IFBenchQwen3.8 27B by 3.975.679.5+3.9
Other results27 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.
- GDPval-AA 2.1Qwen3.8 27B by 800 rating points6271427+800 rating
- SWEBench Pro v2 (Hard)Qwen3.8 27B by 59.614.273.8+59.6
- AutomationBench-AA v1.0.6Qwen3.8 27B by 43.34.948.2+43.3
- AA Agentic IndexQwen3.8 27B by 39.86.746.5+39.8
- AA-OmniscienceQwen3.8 27B by 37.9-47.9-10+37.9
- Misbehavior (Internal) ↓Qwen3.8 27B by 30.127.657.7+30.1
- NL2RepoQwen3.8 27B by 26.815.542.3+26.8
- JobBenchQwen3.8 27B by 25.28.233.4+25.2
- Artificial Analysis Coding IndexQwen3.8 27B by 24.743.468.1+24.7
- BVBQwen3.8 27B by 23.74770.7+23.7
- AA IntelligenceQwen3.8 27B by 1914.733.7+19
- DeepSearchQA (F1)Qwen3.8 27B by 1861.779.7+18
- RealWorldQAQwen3.8 27B by 13.672.385.9+13.6
- OmniDocBench 1.5Qwen3.8 27B by 1180.191.1+11
- GDP (Surge AI)Qwen3.8 27B by 10.4616.4+10.4
- MATH-VisionQwen3.8 27B by 985.694.6+9
- OmniDocBench v1.6Qwen3.8 27B by 982.591.5+9
- Covert misbehavior (Internal) ↓Qwen3.8 27B by 8.924.333.2+8.9
- HLE-DiamondQwen3.8 27B by 8.17.215.3+8.1
- Terminal-Bench 4.0Qwen3.8 27B by 5.605.6+5.6
- ZeroBench (main) originalQwen3.8 27B by 3912+3
- MMMU Pro Standard 10Qwen3.8 27B by 2.973.476.3+2.9
- ComplexConstraints v1.1Gemma 4 31B by 1.531.630.1+1.5
- StrongREJECT (none)tie98.199tie
- Chartography originaltie1212.2tie
- CritPt v2 Hardtie44tie
- Terminal-Bench-Science 0.1tie00tie
Questions people ask
Which is better, Gemma 4 31B or Qwen3.8 27B?
Qwen3.8 27B wins six of the seven areas where both have results: coding, agents, reasoning, images and charts, long documents and following instructions. Gemma 4 31B wins none, but costs 84% less. They are level on facts.
Which is better for coding?
Qwen3.8 27B. It wins 7 of the 7 coding tests both models report; Gemma 4 31B wins none.
Which is cheaper?
Gemma 4 31B costs $0.15 per million input tokens and $0.40 per million output tokens; Qwen3.8 27B costs $0.50 and $3.00. That makes Gemma 4 31B about 84% cheaper for the same work.
How do you compare the two?
We use the 50 benchmark tests both models have published scores on. The verdict counts the 23 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 27 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.