Llama 3.2 3B Instruct vs Qwen2.5 Instruct 72B
Wins 0 of 5 areas
—
Wins 3 of 5 areas
Coding · Long documents · Following instructions
Qwen2.5 Instruct 72B is the stronger all-rounder.Llama 3.2 3B Instruct is cheaper.
Scores updated · 23 tests both models report · How we compare
Where each one wins
Tests won in each of the five 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 software01Qwen2.5 Instruct 72B1 of 1 test
- Long documentsFinding answers in very long texts01Qwen2.5 Instruct 72B1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen2.5 Instruct 72B1 of 1 test
- ReasoningHard problems that need careful thinking11Even1 each
- FactsGetting facts right instead of making them up11Even1 each
Coding, 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.
Llama 3.2 3B Instruct costs 61% 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 Qwen2.5 Instruct 72B pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+23.6points ahead
- Code for real scientific research problemsSciCode+21.5points ahead
- Reasons across sets of long documentsAA-LCR+15.7points ahead
Where Llama 3.2 3B Instruct pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+28points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1.8points ahead
Every test, side by side
All 23 tests both models report. The winning score is in its model's colour; marks a score checked independently.
ReasoningEven
- GPQA DiamondQwen2.5 Instruct 72B by 23.625.549.1+23.6
- Humanity's Last ExamLlama 3.2 3B Instruct by 1.85.33.6+1.8
FactsEven
- AA-Omniscience · Non-hallucinationLlama 3.2 3B Instruct by 2842.514.5+28
- AA-Omniscience · AccuracyQwen2.5 Instruct 72B by 11.26.317.5+11.2
Long documentsQwen2.5 Instruct 72B
- AA-LCRQwen2.5 Instruct 72B by 15.74.320+15.7
Following instructionsQwen2.5 Instruct 72B
- IFBenchQwen2.5 Instruct 72B by 10.726.236.9+10.7
Other results16 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.
- MATHQwen2.5 Instruct 72B by 79.98.588.4+79.9
- HumanEvalQwen2.5 Instruct 72B by 58.62886.6+58.6
- MBPPQwen2.5 Instruct 72B by 39.548.788.2+39.5
- MATH-500 (EM)Qwen2.5 Instruct 72B by 38.841.280+38.8
- MMLU-ProQwen2.5 Instruct 72B by 36.13571.1+36.1
- BBHQwen2.5 Instruct 72B by 3346.879.8+33
- MMMLUQwen2.5 Instruct 72B by 26.947.974.8+26.9
- GSM8KQwen2.5 Instruct 72B by 18.177.795.8+18.1
- MGSMQwen2.5 Instruct 72B by 1858.276.2+18
- MMLUQwen2.5 Instruct 72B by 16.660.477+16.6
- ARC-ChallengeQwen2.5 Instruct 72B by 15.978.694.5+15.9
- HellaSwagQwen2.5 Instruct 72B by 1569.884.8+15
- τ²-Bench Telecom (AA run)Qwen2.5 Instruct 72B by 13.421.134.5+13.4
- IFEvalQwen2.5 Instruct 72B by 6.777.484.1+6.7
- AA-OmniscienceLlama 3.2 3B Instruct by 5.6-47.5-53.1+5.6
- AA IntelligenceQwen2.5 Instruct 72B by 25.77.7+2
Questions people ask
Which is better, Llama 3.2 3B Instruct or Qwen2.5 Instruct 72B?
Qwen2.5 Instruct 72B wins three of the five areas where both have results: coding, long documents and following instructions. Llama 3.2 3B Instruct wins none, but costs 61% less. They are level on reasoning and facts.
Which is better for coding?
Qwen2.5 Instruct 72B. It wins the one coding test both models report.
Which is cheaper?
Llama 3.2 3B Instruct costs $0.05 per million input tokens and $0.33 per million output tokens; Qwen2.5 Instruct 72B costs $0.47 and $0.49. That makes Llama 3.2 3B Instruct about 61% cheaper for the same work.
How do you compare the two?
We use the 23 benchmark tests both models have published scores on. The verdict counts the 7 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 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.