MiniMax M3 vs Mistral Medium 3.5 128B
Wins 6 of 7 areas
Coding · Agents · Reasoning · Images and charts · Long documents · Following instructions
Wins 0 of 7 areas
—
MiniMax M3 is the stronger all-rounder.
Scores updated · 25 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 software50MiniMax M35 of 5 tests
- AgentsCarrying out multi-step tasks on its own30MiniMax M33 of 4 tests · 1 tie
- ReasoningHard problems that need careful thinking30MiniMax M33 of 3 tests
- Images and chartsUnderstanding pictures, charts and video20MiniMax M32 of 2 tests
- Long documentsFinding answers in very long texts10MiniMax M31 of 1 test
- Following instructionsDoing exactly what it is asked10MiniMax M31 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
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.
MiniMax M3 costs 83% 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 MiniMax M3 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+63.2points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+34.9points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+25.2points ahead
Where Mistral Medium 3.5 128B pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+8points ahead
Every test, side by side
All 25 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingMiniMax M3
- LMArena · WebDevMiniMax M3 by 217 rating points14821265+217 rating
- Terminal-Bench 2.1MiniMax M3 by 14.665.250.6+14.6
- Terminal-Bench HardMiniMax M3 by 9.142.433.3+9.1
- SciCodeMiniMax M3 by 6.947.140.2+6.9
- SWE-bench VerifiedMiniMax M3 by 2.980.577.6+2.9
AgentsMiniMax M3
- BrowseCompMiniMax M3 by 34.983.548.6+34.9
- GDPValMiniMax M3 by 24.137.313.2+24.1
- Terminal-Bench 4.0MiniMax M3 by 220+2
- τ-Bench V3 · Bankingtie15.315.1tie
ReasoningMiniMax M3
- Humanity's Last ExamMiniMax M3 by 25.23913.8+25.2
- GPQA DiamondMiniMax M3 by 18.192.974.8+18.1
- CritPtMiniMax M3 by 3.73.70+3.7
FactsEven
- AA-Omniscience · Non-hallucinationMiniMax M3 by 63.281.618.4+63.2
- AA-Omniscience · AccuracyMistral Medium 3.5 128B by 816.724.7+8
Images and chartsMiniMax M3
- MMMU-ProMiniMax M3 by 13.778.664.9+13.7
- LMArena · VisionMiniMax M3 by 32 rating points12541222+32 rating
Following instructionsMiniMax M3
- IFBenchMiniMax M3 by 14.182.968.8+14.1
Other results7 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-OmniscienceMiniMax M3 by 38.21.4-36.8+38.2
- AA Agentic IndexMiniMax M3 by 21.430.89.4+21.4
- Finance Agent v2MiniMax M3 by 16.248.332.1+16.2
- AA IntelligenceMiniMax M3 by 1529.214.2+15
- Artificial Analysis Coding IndexMiniMax M3 by 11.758.646.9+11.7
- τ²-Bench Telecom (AA run)Mistral Medium 3.5 128B by 5.388.994.2+5.3
- τ³-Bench Bankingtie1313.4tie
Questions people ask
Which is better, MiniMax M3 or Mistral Medium 3.5 128B?
MiniMax M3 wins six of the seven areas where both have results: coding, agents, reasoning, images and charts, long documents and following instructions. Mistral Medium 3.5 128B wins none. They are level on facts.
Which is better for coding?
MiniMax M3. It wins 5 of the 5 coding tests both models report; Mistral Medium 3.5 128B wins none.
Which is cheaper?
MiniMax M3 costs $0.30 per million input tokens and $1.20 per million output tokens; Mistral Medium 3.5 128B costs $1.50 and $7.50. That makes MiniMax M3 about 83% cheaper for the same work.
How do you compare the two?
We use the 25 benchmark tests both models have published scores on. The verdict counts the 18 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 7 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.