Gemma 4 26B A4B vs Gemma 4 31B
Wins 1 of 8 areas
Math
Wins 6 of 8 areas
Coding · Agents · Reasoning · Images and charts · Long documents · Following instructions
Gemma 4 31B is the stronger all-rounder.Gemma 4 26B A4B is cheaper and better at math.
Scores updated · 73 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.
- CodingWriting and fixing software07Gemma 4 31B7 of 8 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own04Gemma 4 31B4 of 4 tests
- ReasoningHard problems that need careful thinking03Gemma 4 31B3 of 3 tests
- Images and chartsUnderstanding pictures, charts and video13Gemma 4 31B3 of 5 tests · 1 tie
- Long documentsFinding answers in very long texts02Gemma 4 31B2 of 2 tests
- Following instructionsDoing exactly what it is asked01Gemma 4 31B1 of 1 test
- MathCompetition and research-level math10Gemma 4 26B A4B1 of 3 tests · 2 ties
- FactsGetting facts right instead of making them up11Even1 each · 1 tie
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 26B A4B costs 6% 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 Gemma 4 31B pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+34.6points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+34.4points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+22.8points ahead
Where Gemma 4 26B A4B pulls ahead
- Harvard-MIT high-school math contest problemsHMMT Feb. 2026+1.8points ahead
- Reasoning about charts from research papersCharXiv (RQ)+1.1points ahead
Every test, side by side
All 73 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemma 4 31B
- SWE-bench VerifiedGemma 4 31B by 34.617.452+34.6
- SWE-bench MultilingualGemma 4 31B by 34.417.351.7+34.4
- Terminal-Bench HardGemma 4 31B by 22.813.636.4+22.8
- SWE-bench ProGemma 4 31B by 21.913.835.7+21.9
- SciCodeGemma 4 31B by 5.54045.5+5.5
- Terminal-Bench 2.1Gemma 4 31B by 4.43943.4+4.4
- LiveCodeBench v6Gemma 4 31B by 2.977.180+2.9
- LMArena · WebDevtie13591365tie
AgentsGemma 4 31B
- AA IT-Bench SREGemma 4 31B by 13.723.637.3+13.7
- MCP AtlasGemma 4 31B by 7.25057.2+7.2
- τ-Bench V3 · BankingGemma 4 31B by 2.81214.8+2.8
- GDPValGemma 4 31B by 2.73.46.1+2.7
ReasoningGemma 4 31B
- GPQA DiamondGemma 4 31B by 6.579.285.7+6.5
- Humanity's Last ExamGemma 4 31B by 4.319.323.6+4.3
- CritPtGemma 4 31B by 1.401.4+1.4
FactsEven
- Vectara HHEM hallucination ratelower is betterGemma 4 26B A4B by 2.25.27.4+2.2
- AA-Omniscience · Non-hallucinationGemma 4 31B by 1.413.615+1.4
- AA-Omniscience · Accuracytie19.120tie
Images and chartsGemma 4 31B
- MMMU-ProGemma 4 31B by 4.269.273.4+4.2
- MMMUGemma 4 31B by 278.480.4+2
- LMArena · VisionGemma 4 31B by 17 rating points12601277+17 rating
- CharXiv (RQ)Gemma 4 26B A4B by 1.16967.9+1.1
- MathVistatie79.479.3tie
MathGemma 4 26B A4B
- HMMT Feb. 2026Gemma 4 26B A4B by 1.87977.2+1.8
- AIME 2026tie88.389.2tie
- IMOAnswerBenchtie74.374.5tie
Long documentsGemma 4 31B
- MRCR v2 8 needle 128k (average)Gemma 4 31B by 22.344.166.4+22.3
- AA-LCRGemma 4 31B by 465.769.7+4
Following instructionsGemma 4 31B
- IFBenchGemma 4 31B by 3.272.475.6+3.2
Other results44 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.
- Codeforces ELOGemma 4 31B by 432 rating points17182150+432 rating
- τ²-Bench Telecom (AA run)Gemma 4 31B by 16.343.659.9+16.3
- SkillsBench Avg5Gemma 4 31B by 11.312.323.6+11.3
- Claw-Eval AvgGemma 4 26B A4B by 10.358.848.5+10.3
- BIG-Bench Extra HardGemma 4 31B by 9.664.874.4+9.6
- HLE (with tools)Gemma 4 31B by 9.317.226.5+9.3
- Tool-DecathlonGemma 4 31B by 9.21221.2+9.2
- Terminal-Bench 2.0Gemma 4 31B by 8.734.242.9+8.7
- vectara_avg_summary_lengthGemma 4 31B by 8.767.175.8+8.7
- τ²-BenchGemma 4 31B by 8.768.276.9+8.7
- τ³-BenchGemma 4 31B by 8.55967.5+8.5
- DeepPlanningGemma 4 31B by 7.816.224+7.8
- VITA-BenchGemma 4 31B by 6.136.943+6.1
- OmniDocBench 1.5Gemma 4 31B by 5.774.480.1+5.7
- AA Agentic IndexGemma 4 26B A4B by 4.3116.7+4.3
- SuperGPQAGemma 4 31B by 4.361.465.7+4.3
- Artificial Analysis Coding IndexGemma 4 31B by 4.139.343.4+4.1
- MCPMarkGemma 4 31B by 3.914.218.1+3.9
- NL2RepoGemma 4 31B by 3.911.615.5+3.9
- MATH-VisionGemma 4 31B by 3.282.485.6+3.2
- MedXPertQA MMGemma 4 31B by 3.258.161.3+3.2
- WideSearchGemma 4 26B A4B by 3.138.335.2+3.1
- Claw Eval (pass@3)Gemma 4 26B A4B by 32825+3
- HMMT Feb. 2025Gemma 4 26B A4B by 391.788.7+3
- QwenClawBenchGemma 4 31B by 338.741.7+3
- AA-OmniscienceGemma 4 31B by 2.9-50.8-47.9+2.9
- MMLU-ProGemma 4 31B by 2.682.685.2+2.6
- vectara_factual_consistencyGemma 4 26B A4B by 2.294.892.6+2.2
- MMMLUGemma 4 31B by 2.186.388.4+2.1
- AA IntelligenceGemma 4 26B A4B by 216.714.7+2
- GPQA (unspecified)Gemma 4 31B by 282.384.3+2
- MMBench EN-DEV-v1.1Gemma 4 31B by 1.98990.9+1.9
- QwenWebBenchGemma 4 31B by 19 rating points11781197+19 rating
- HallusionBenchGemma 4 31B by 1.366.167.4+1.3
- CC-OCRGemma 4 31B by 1.274.575.7+1.2
- MMLU-ReduxGemma 4 31B by 192.793.7+1
- t2-benchtie85.586.4tie
- AI2Dtie88.389tie
- SimpleVQAtie52.252.9tie
- vectara_answer_ratetie99.8100tie
- C-Evaltie82.582.6tie
- RealWorldQAtie72.272.3tie
- HMMT Nov. 2025tie87.587.5tie
- OmniDocBench 1.5 (average edit distance, lower is better)lower is bettertie0.10.1tie
Questions people ask
Which is better, Gemma 4 26B A4B or Gemma 4 31B?
Gemma 4 31B wins six of the eight areas we test: coding, agents, reasoning, images and charts, long documents and following instructions. Gemma 4 26B A4B wins math, and costs 6% less. They are level on facts.
Which is better for coding?
Gemma 4 31B. It wins 7 of the 8 coding tests both models report; Gemma 4 26B A4B wins none, and 1 is a tie.
Which is cheaper?
Gemma 4 26B A4B costs $0.13 per million input tokens and $0.40 per million output tokens; Gemma 4 31B costs $0.17 and $0.40. That makes Gemma 4 26B A4B about 6% cheaper for the same work.
How do you compare the two?
We use the 73 benchmark tests both models have published scores on. The verdict counts the 29 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 44 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.