Ministral 3 8B vs Qwen3 32B
Wins 1 of 6 areas
Long documents
Wins 5 of 6 areas
Coding · Agents · Reasoning · Facts · Following instructions
Qwen3 32B is the stronger all-rounder.Ministral 3 8B is cheaper and better at long documents.
Scores updated · 25 tests both models report · How we compare
Where each one wins
Tests won in each of the six 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.
- FactsGetting facts right instead of making them up03Qwen3 32B3 of 3 tests
- ReasoningHard problems that need careful thinking02Qwen3 32B2 of 3 tests · 1 tie
- CodingWriting and fixing software12Qwen3 32B2 of 3 tests
- AgentsCarrying out multi-step tasks on its own01Qwen3 32B1 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked01Qwen3 32B1 of 1 test
- Long documentsFinding answers in very long texts10Ministral 3 8B1 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.
Ministral 3 8B costs 63% 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 Qwen3 32B pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+19.7points ahead
- Code for real scientific research problemsSciCode+15.3points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+12.1points ahead
Where Ministral 3 8B pulls ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+1.5points 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.
CodingQwen3 32B
- SciCodeQwen3 32B by 15.320.736+15.3
- Terminal-Bench HardMinistral 3 8B by 1.54.53+1.5
- Terminal-Bench 2.1Qwen3 32B by 1.14.15.2+1.1
AgentsQwen3 32B
- τ-Bench V3 · BankingQwen3 32B by 1.73.75.4+1.7
- GDPValtie00tie
- Terminal-Bench 4.0tie00tie
ReasoningQwen3 32B
- GPQA DiamondQwen3 32B by 19.747.166.8+19.7
- Humanity's Last ExamQwen3 32B by 3.14.37.4+3.1
- CritPttie00.3tie
FactsQwen3 32B
- Vectara HHEM hallucination ratelower is betterQwen3 32B by 15.821.75.9+15.8
- AA-Omniscience · Non-hallucinationQwen3 32B by 12.15.817.9+12.1
- AA-Omniscience · AccuracyQwen3 32B by 4.41317.4+4.4
Following instructionsQwen3 32B
- IFBenchQwen3 32B by 7.229.136.3+7.2
Other results11 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.
- Arena HardQwen3 32B by 42.950.993.8+42.9
- AA-OmniscienceQwen3 32B by 18.5-69-50.4+18.5
- vectara_factual_consistencyQwen3 32B by 15.878.394.1+15.8
- AIME 2025Ministral 3 8B by 5.878.772.9+5.8
- Artificial Analysis Coding IndexQwen3 32B by 5.69.715.3+5.6
- AIME 2024Ministral 3 8B by 4.68681.4+4.6
- LiveCodeBenchQwen3 32B by 4.161.665.7+4.1
- τ²-Bench Telecom (AA run)Qwen3 32B by 3.226.629.8+3.2
- AA IntelligenceQwen3 32B by 3.15.58.6+3.1
- vectara_avg_summary_lengthMinistral 3 8B by 23.6 rating points139.4115.8+23.6 rating
- vectara_answer_ratetie99.199.9tie
Questions people ask
Which is better, Ministral 3 8B or Qwen3 32B?
Qwen3 32B wins five of the six areas where both have results: coding, agents, reasoning, facts and following instructions. Ministral 3 8B wins long documents, and costs 63% less.
Which is better for coding?
Qwen3 32B. It wins 2 of the 3 coding tests both models report; Ministral 3 8B wins 1.
Which is cheaper?
Ministral 3 8B costs $0.15 per million input tokens and $0.15 per million output tokens; Qwen3 32B costs $0.16 and $0.64. That makes Ministral 3 8B about 63% 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 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 11 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.