Gemma 4 31B vs GPT-5.6 Luna
Wins 0 of 7 areas
—
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
GPT-5.6 Luna is the stronger all-rounder.Gemma 4 31B is cheaper.
Scores updated · 34 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 software05GPT-5.6 Luna5 of 5 tests
- AgentsCarrying out multi-step tasks on its own05GPT-5.6 Luna5 of 5 tests
- ReasoningHard problems that need careful thinking03GPT-5.6 Luna3 of 3 tests
- MathCompetition and research-level math02GPT-5.6 Luna2 of 2 tests
- Long documentsFinding answers in very long texts02GPT-5.6 Luna2 of 2 tests
- FactsGetting facts right instead of making them up12GPT-5.6 Luna2 of 3 tests
- Images and chartsUnderstanding pictures, charts and video22Even2 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 61% 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 GPT-5.6 Luna pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+46.9points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+46.3points ahead
- Real work tasks from 44 professionsGDPVal+42points ahead
Where Gemma 4 31B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+7.6points ahead
- Math problems shown in pictures and chartsMathVista+2.4points ahead
Every test, side by side
All 34 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.6 Luna
- SWE-bench VerifiedGPT-5.6 Luna by 46.946.193+46.9
- Terminal-Bench 2.1GPT-5.6 Luna by 37.543.480.9+37.5
- SWE-bench ProGPT-5.6 Luna by 2735.762.7+27
- LMArena · WebDevGPT-5.6 Luna by 154 rating points13651519+154 rating
- SciCodeGPT-5.6 Luna by 8.145.553.6+8.1
AgentsGPT-5.6 Luna
- BrowseCompGPT-5.6 Luna by 46.33783.3+46.3
- GDPValGPT-5.6 Luna by 42648+42
- τ-Bench V3 · BankingGPT-5.6 Luna by 16.314.831.1+16.3
- Terminal-Bench 4.0GPT-5.6 Luna by 11.6011.6+11.6
- AA IT-Bench SREGPT-5.6 Luna by 337.340.3+3
ReasoningGPT-5.6 Luna
- CritPtGPT-5.6 Luna by 19.21.420.6+19.2
- Humanity's Last ExamGPT-5.6 Luna by 15.923.639.5+15.9
- GPQA DiamondGPT-5.6 Luna by 5.485.791.1+5.4
FactsGPT-5.6 Luna
- SimpleQA VerifiedGPT-5.6 Luna by 30.610.441+30.6
- AA-Omniscience · AccuracyGPT-5.6 Luna by 22.72042.7+22.7
- AA-Omniscience · Non-hallucinationGemma 4 31B by 7.6157.4+7.6
Images and chartsEven
- CharXiv (RQ)GPT-5.6 Luna by 14.867.982.7+14.8
- MMMU-ProGPT-5.6 Luna by 5.273.478.6+5.2
- MathVistaGemma 4 31B by 2.479.376.9+2.4
- LMArena · VisionGemma 4 31B by 23 rating points12791256+23 rating
MathGPT-5.6 Luna
- HMMT Feb. 2026GPT-5.6 Luna by 21.377.298.5+21.3
- AIME 2026GPT-5.6 Luna by 8.489.297.6+8.4
Long documentsGPT-5.6 Luna
- AA-LCRGPT-5.6 Luna by 1469.783.7+14
- MRCR v2 8 needle 128k (average)GPT-5.6 Luna by 8.466.474.8+8.4
Other results10 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-OmniscienceGPT-5.6 Luna by 37.6-47.9-10.3+37.6
- AA Agentic IndexGPT-5.6 Luna by 366.742.7+36
- Artificial Analysis Coding IndexGPT-5.6 Luna by 2843.471.4+28
- AA IntelligenceGPT-5.6 Luna by 22.614.737.3+22.6
- HLE (with tools)GPT-5.6 Luna by 22.426.548.9+22.4
- Terminal-Bench 4.0GPT-5.6 Luna by 17.3017.3+17.3
- GDP (Surge AI)GPT-5.6 Luna by 16.7622.7+16.7
- HallusionBenchGemma 4 31B by 8.467.459+8.4
- SimpleVQAtie52.953.2tie
- RealWorldQAtie72.372.4tie
Questions people ask
Which is better, Gemma 4 31B or GPT-5.6 Luna?
GPT-5.6 Luna wins six of the seven areas where both have results: coding, agents, reasoning, facts, math and long documents. Gemma 4 31B wins none, but costs 61% less. They are level on images and charts.
Which is better for coding?
GPT-5.6 Luna. It wins 5 of the 5 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; GPT-5.6 Luna costs $0.20 and $1.20. That makes Gemma 4 31B about 61% cheaper for the same work.
How do you compare the two?
We use the 34 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 10 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.