Ministral 3 14B vs Qwen3 14B
Wins 2 of 6 areas
Agents · Long documents
Wins 3 of 6 areas
Reasoning · Facts · Following instructions
Qwen3 14B wins more areas, narrowly.Ministral 3 14B is cheaper and better at agents.
Scores updated · 26 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 14B3 of 3 tests
- ReasoningHard problems that need careful thinking01Qwen3 14B1 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked01Qwen3 14B1 of 1 test
- AgentsCarrying out multi-step tasks on its own10Ministral 3 14B1 of 3 tests · 2 ties
- Long documentsFinding answers in very long texts10Ministral 3 14B1 of 1 test
- CodingWriting and fixing software11Even1 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.
Ministral 3 14B costs 77% 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 14B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+16.7points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+8.5points ahead
- Code for real scientific research problemsSciCode+6.9points ahead
Where Ministral 3 14B pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+4.8points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+1points ahead
Every test, side by side
All 26 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- SciCodeQwen3 14B by 6.923.830.7+6.9
- Terminal-Bench 2.1Ministral 3 14B by 4.89.74.9+4.8
- Terminal-Bench Hardtie4.53.8tie
AgentsMinistral 3 14B
- τ-Bench V3 · BankingMinistral 3 14B by 16.65.6+1
- GDPValtie00tie
- Terminal-Bench 4.0tie00tie
ReasoningQwen3 14B
- GPQA DiamondQwen3 14B by 3.257.260.4+3.2
- Humanity's Last Examtie4.64.5tie
- CritPttie00tie
FactsQwen3 14B
- AA-Omniscience · Non-hallucinationQwen3 14B by 16.77.524.2+16.7
- Vectara HHEM hallucination ratelower is betterQwen3 14B by 1419.45.4+14
- AA-Omniscience · AccuracyQwen3 14B by 1.713.615.3+1.7
Other results12 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-OmniscienceQwen3 14B by 17.5-66.4-49+17.5
- AIME 2025Ministral 3 14B by 14.68570.4+14.6
- vectara_factual_consistencyQwen3 14B by 1480.694.6+14
- Arena HardMinistral 3 14B by 12.455.142.7+12.4
- τ²-Bench Telecom (AA run)Qwen3 14B by 7.327.234.5+7.3
- AIME 2024Ministral 3 14B by 6.189.883.7+6.1
- LiveCodeBenchMinistral 3 14B by 5.364.659.3+5.3
- WildBenchMinistral 3 14B by 3.468.565.1+3.4
- vectara_avg_summary_lengthMinistral 3 14B by 24.7 rating points135.8111.1+24.7 rating
- AA IntelligenceQwen3 14B by 2.268.2+2.2
- Artificial Analysis Coding Indextie14.413.8tie
- vectara_answer_ratetie99.699.9tie
Questions people ask
Which is better, Ministral 3 14B or Qwen3 14B?
Qwen3 14B wins three of the six areas where both have results: reasoning, facts and following instructions. Ministral 3 14B wins agents and long documents, and costs 77% less. They are level on coding.
Which is better for coding?
Neither. They win 1 coding test each of the 3 both models report, and 1 is a tie.
Which is cheaper?
Ministral 3 14B costs $0.20 per million input tokens and $0.20 per million output tokens; Qwen3 14B costs $0.35 and $1.40. That makes Ministral 3 14B about 77% cheaper for the same work.
How do you compare the two?
We use the 26 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 12 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.