Gemma 4 31B vs Qwen3.5 122B A10B
Wins 1 of 7 areas
Facts
Wins 5 of 7 areas
Coding · Agents · Reasoning · Images and charts · Long documents
Qwen3.5 122B A10B is the stronger all-rounder.Gemma 4 31B is cheaper and better at facts.
Scores updated · 49 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.
- Images and chartsUnderstanding pictures, charts and video14Qwen3.5 122B A10B4 of 5 tests
- CodingWriting and fixing software34Qwen3.5 122B A10B4 of 8 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own01Qwen3.5 122B A10B1 of 3 tests · 2 ties
- ReasoningHard problems that need careful thinking01Qwen3.5 122B A10B1 of 3 tests · 2 ties
- Long documentsFinding answers in very long texts01Qwen3.5 122B A10B1 of 1 test
- FactsGetting facts right instead of making them up21Gemma 4 31B2 of 3 tests
- 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 84% 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 122B A10B pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+20points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+10.3points ahead
- Real work tasks from 44 professionsGDPVal+9.7points ahead
Where Gemma 4 31B pulls ahead
- Code for real scientific research problemsSciCode+5.8points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+5.3points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+2.1points ahead
Every test, side by side
All 49 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.5 122B A10B
- SWE-bench VerifiedQwen3.5 122B A10B by 205272+20
- SWE-bench MultilingualQwen3.5 122B A10B by 10.351.762+10.3
- SWE-bench ProQwen3.5 122B A10B by 9.235.744.9+9.2
- SciCodeGemma 4 31B by 5.845.539.7+5.8
- Terminal-Bench HardGemma 4 31B by 5.336.431.1+5.3
- Terminal-Bench 2.1Qwen3.5 122B A10B by 4.243.447.6+4.2
- LiveCodeBench v6Gemma 4 31B by 1.18078.9+1.1
- LMArena · WebDevtie13651360tie
AgentsQwen3.5 122B A10B
- GDPValQwen3.5 122B A10B by 9.7615.7+9.7
- τ-Bench V3 · Bankingtie14.815.3tie
- Terminal-Bench 4.0tie00tie
ReasoningQwen3.5 122B A10B
- Humanity's Last ExamQwen3.5 122B A10B by 1.623.625.2+1.6
- CritPttie1.40.6tie
- GPQA Diamondtie85.785.7tie
FactsGemma 4 31B
- AA-Omniscience · AccuracyQwen3.5 122B A10B by 4.42024.4+4.4
- Vectara HHEM hallucination ratelower is betterGemma 4 31B by 3.87.411.2+3.8
- AA-Omniscience · Non-hallucinationGemma 4 31B by 2.11512.9+2.1
Images and chartsQwen3.5 122B A10B
- CharXiv (RQ)Qwen3.5 122B A10B by 9.367.977.2+9.3
- MathVistaQwen3.5 122B A10B by 8.179.387.4+8.1
- MMMUQwen3.5 122B A10B by 3.580.483.9+3.5
- LMArena · VisionGemma 4 31B by 32 rating points12771245+32 rating
- MMMU-ProQwen3.5 122B A10B by 1.673.475+1.6
Long documentsQwen3.5 122B A10B
- AA-LCRQwen3.5 122B A10B by 6.669.776.3+6.6
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.
- τ²-Bench Telecom (AA run)Qwen3.5 122B A10B by 33.759.993.6+33.7
- WideSearchQwen3.5 122B A10B by 25.335.260.5+25.3
- RealWorldQAQwen3.5 122B A10B by 12.872.385.1+12.8
- vectara_avg_summary_lengthQwen3.5 122B A10B by 10.675.886.4+10.6
- OmniDocBench 1.5Qwen3.5 122B A10B by 9.780.189.8+9.7
- VITA-BenchGemma 4 31B by 9.44333.6+9.4
- C-EvalQwen3.5 122B A10B by 9.382.691.9+9.3
- SimpleVQAQwen3.5 122B A10B by 8.852.961.7+8.8
- t2-benchGemma 4 31B by 6.986.479.5+6.9
- Terminal-Bench 2.0Qwen3.5 122B A10B by 6.542.949.4+6.5
- AA-OmniscienceQwen3.5 122B A10B by 6.4-47.9-41.5+6.4
- CC-OCRQwen3.5 122B A10B by 6.175.781.8+6.1
- AI2DQwen3.5 122B A10B by 4.38993.3+4.3
- vectara_factual_consistencyGemma 4 31B by 3.892.688.8+3.8
- AA Agentic IndexQwen3.5 122B A10B by 2.96.79.6+2.9
- Artificial Analysis Coding IndexQwen3.5 122B A10B by 2.343.445.7+2.3
- MMMLUGemma 4 31B by 1.788.486.7+1.7
- MMLU-ProQwen3.5 122B A10B by 1.585.286.7+1.5
- SuperGPQAQwen3.5 122B A10B by 1.465.767.1+1.4
- AA Intelligencetie14.715.6tie
- MATH-Visiontie85.686.2tie
- MMLU-Reduxtie93.794tie
- HallusionBenchtie67.467.6tie
- vectara_answer_ratetie10099.8tie
- DeepPlanningtie2424.1tie
Questions people ask
Which is better, Gemma 4 31B or Qwen3.5 122B A10B?
Qwen3.5 122B A10B wins five of the seven areas where both have results: coding, agents, reasoning, images and charts and long documents. Gemma 4 31B wins facts, and costs 84% less. They are level on following instructions.
Which is better for coding?
Qwen3.5 122B A10B. It wins 4 of the 8 coding tests both models report; Gemma 4 31B wins 3, and 1 is a tie.
Which is cheaper?
Gemma 4 31B costs $0.17 per million input tokens and $0.40 per million output tokens; Qwen3.5 122B A10B costs $0.40 and $3.20. That makes Gemma 4 31B about 84% cheaper for the same work.
How do you compare the two?
We use the 49 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 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.