Ministral 3 3B vs Qwen3.5 2B
Wins 1 of 7 areas
Agents
Wins 5 of 7 areas
Coding · Facts · Images and charts · Long documents · Following instructions
Qwen3.5 2B is the stronger all-rounder.Ministral 3 3B is better at agents.
Scores updated · 32 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 video02Qwen3.5 2B2 of 2 tests
- CodingWriting and fixing software12Qwen3.5 2B2 of 3 tests
- FactsGetting facts right instead of making them up01Qwen3.5 2B1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts01Qwen3.5 2B1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3.5 2B1 of 1 test
- AgentsCarrying out multi-step tasks on its own10Ministral 3 3B1 of 2 tests · 1 tie
- ReasoningHard problems that need careful thinking11Even1 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.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3.5 2B pulls ahead
- Reads text in imagesOCRBench+12.6points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+9.8points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+6points ahead
Where Ministral 3 3B pulls ahead
- Code for real scientific research problemsSciCode+12.5points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+2.8points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+2.2points ahead
Every test, side by side
All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.5 2B
- SciCodeMinistral 3 3B by 12.515.32.8+12.5
- Terminal-Bench HardQwen3.5 2B by 3.803.8+3.8
- Terminal-Bench 2.1Qwen3.5 2B by 303+3
AgentsMinistral 3 3B
- τ-Bench V3 · BankingMinistral 3 3B by 2.24.72.5+2.2
- GDPValtie00tie
ReasoningEven
- GPQA DiamondQwen3.5 2B by 9.835.845.6+9.8
- Humanity's Last ExamMinistral 3 3B by 2.85.42.6+2.8
- CritPttie00tie
FactsQwen3.5 2B
- AA-Omniscience · Non-hallucinationQwen3.5 2B by 619.825.8+6
- AA-Omniscience · Accuracytie98.1tie
Images and chartsQwen3.5 2B
Following instructionsQwen3.5 2B
- IFBenchQwen3.5 2B by 4.726.831.5+4.7
Other results18 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.
- AIME 2024Ministral 3 3B by 60.577.517+60.5
- AIME 2025Ministral 3 3B by 53.172.119+53.1
- τ²-Bench Telecom (AA run)Qwen3.5 2B by 44.124.969+44.1
- MMLU-Pro testMinistral 3 3B by 17.747.529.8+17.7
- MMLU testMinistral 3 3B by 11.76654.3+11.7
- MMBench DEV_EN_V11Qwen3.5 2B by 6.869.276+6.8
- OCRBench v2_enQwen3.5 2B by 6.741.448.1+6.7
- AA-OmniscienceQwen3.5 2B by 3.8-64-60.2+3.8
- MMMU DEV_VALMinistral 3 3B by 3.450.847.4+3.4
- DocVQA-valQwen3.5 2B by 389.692.6+3
- AA IntelligenceQwen3.5 2B by 2.14.86.9+2.1
- Artificial Analysis Coding IndexMinistral 3 3B by 1.94.82.9+1.9
- MMMBQwen3.5 2B by 1.173.474.5+1.1
- ChartQA Testtie79.178.4tie
- GQA TestDev_Balancedtie54.453.9tie
- MTL MMBench_DEVtie67.466.9tie
- HallusionBenchtie65.265.5tie
- MMBench_DEVtie0.70.7tie
Questions people ask
Which is better, Ministral 3 3B or Qwen3.5 2B?
Qwen3.5 2B wins five of the seven areas where both have results: coding, facts, images and charts, long documents and following instructions. Ministral 3 3B wins agents. They are level on reasoning.
Which is better for coding?
Qwen3.5 2B. It wins 2 of the 3 coding tests both models report; Ministral 3 3B wins 1.
How do you compare the two?
We use the 32 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 18 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.