Gemma 4 31B vs Mistral Medium 3.1 (Non-Reasoning)
Wins 7 of 7 areas
Coding · Agents · Reasoning · Facts · Images and charts · Long documents · Following instructions
Wins 0 of 7 areas
—
Gemma 4 31B is the stronger all-rounder.
Scores updated · 24 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 software30Gemma 4 31B3 of 3 tests
- ReasoningHard problems that need careful thinking30Gemma 4 31B3 of 3 tests
- AgentsCarrying out multi-step tasks on its own20Gemma 4 31B2 of 3 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video20Gemma 4 31B2 of 2 tests
- FactsGetting facts right instead of making them up10Gemma 4 31B1 of 3 tests · 2 ties
- Long documentsFinding answers in very long texts10Gemma 4 31B1 of 1 test
- Following instructionsDoing exactly what it is asked10Gemma 4 31B1 of 1 test
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 76% 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
- Follows unfamiliar, precisely checkable instructionsIFBench+35.8points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+29.5points ahead
- Reasons across sets of long documentsAA-LCR+27points ahead
Where Mistral Medium 3.1 (Non-Reasoning) pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 24 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemma 4 31B
- Terminal-Bench 2.1Gemma 4 31B by 29.543.413.9+29.5
- Terminal-Bench HardGemma 4 31B by 25.836.410.6+25.8
- SciCodeGemma 4 31B by 13.145.532.4+13.1
AgentsGemma 4 31B
- τ-Bench V3 · BankingGemma 4 31B by 6.614.88.2+6.6
- GDPValGemma 4 31B by 6.16.10+6.1
- Terminal-Bench 4.0tie00tie
ReasoningGemma 4 31B
- GPQA DiamondGemma 4 31B by 26.985.758.8+26.9
- Humanity's Last ExamGemma 4 31B by 18.923.64.7+18.9
- CritPtGemma 4 31B by 1.41.40+1.4
FactsGemma 4 31B
- Vectara HHEM hallucination ratelower is betterGemma 4 31B by 15.37.422.7+15.3
- AA-Omniscience · Accuracytie2020.8tie
- AA-Omniscience · Non-hallucinationtie1515.4tie
Images and chartsGemma 4 31B
- MMMU-ProGemma 4 31B by 19.273.454.2+19.2
- LMArena · VisionGemma 4 31B by 105 rating points12771172+105 rating
Following instructionsGemma 4 31B
- IFBenchGemma 4 31B by 35.875.639.8+35.8
Other results8 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.
- Artificial Analysis Coding IndexGemma 4 31B by 22.943.420.5+22.9
- τ²-Bench Telecom (AA run)Gemma 4 31B by 19.359.940.6+19.3
- vectara_factual_consistencyGemma 4 31B by 15.392.677.3+15.3
- vectara_avg_summary_lengthMistral Medium 3.1 (Non-Reasoning) by 67.1 rating points75.8142.9+67.1 rating
- AA IntelligenceGemma 4 31B by 5.514.79.2+5.5
- AA Agentic IndexGemma 4 31B by 3.66.73.1+3.6
- AA-OmniscienceMistral Medium 3.1 (Non-Reasoning) by 1.6-47.9-46.3+1.6
- vectara_answer_ratetie10099.7tie
Questions people ask
Which is better, Gemma 4 31B or Mistral Medium 3.1 (Non-Reasoning)?
Gemma 4 31B wins all seven areas where both have results: coding, agents, reasoning, facts, images and charts, long documents and following instructions. Mistral Medium 3.1 (Non-Reasoning) wins none.
Which is better for coding?
Gemma 4 31B. It wins 3 of the 3 coding tests both models report; Mistral Medium 3.1 (Non-Reasoning) wins none.
Which is cheaper?
Gemma 4 31B costs $0.17 per million input tokens and $0.40 per million output tokens; Mistral Medium 3.1 (Non-Reasoning) costs $0.40 and $2.00. That makes Gemma 4 31B about 76% cheaper for the same work.
How do you compare the two?
We use the 24 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 8 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.