GPT-4o is the stronger all-rounder.Qwen2.5 Instruct 72B is cheaper.
Scores updated · 62 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 software30GPT-4o3 of 3 tests
- FactsGetting facts right instead of making them up20GPT-4o2 of 2 tests
- Long documentsFinding answers in very long texts20GPT-4o2 of 2 tests
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
- Following instructionsDoing exactly what it is asked00Even0 each · 1 tie
Following instructions rests on a single test.
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.
Qwen2.5 Instruct 72B costs 95% 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 GPT-4o pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+47.6points ahead
- Reasons across sets of long documentsAA-LCR+36points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+9.4points ahead
Where Qwen2.5 Instruct 72B pulls ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1.8points ahead
Every test, side by side
All 62 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-4o
- SWE-bench VerifiedGPT-4o by 9.433.223.8+9.4
- SciCodeGPT-4o by 6.633.326.7+6.6
- Terminal-Bench HardGPT-4o by 3.88.34.5+3.8
ReasoningEven
- GPQA DiamondGPT-4o by 3.552.649.1+3.5
- Humanity's Last ExamQwen2.5 Instruct 72B by 1.81.83.6+1.8
- CritPttie00tie
FactsGPT-4o
- AA-Omniscience · Non-hallucinationGPT-4o by 47.662.114.5+47.6
- AA-Omniscience · AccuracyGPT-4o by 6.223.717.5+6.2
Long documentsGPT-4o
- AA-LCRGPT-4o by 365620+36
- LongBench v2GPT-4o by 8.748.139.4+8.7
Other results51 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.
- CodeforcesGPT-4o by 734.2 rating points75924.8+734.2 rating
- AA-OmniscienceGPT-4o by 42.6-10.5-53.1+42.6
- SimpleQAGPT-4o by 29.138.29.1+29.1
- LiveCodeBenchQwen2.5 Instruct 72B by 17.238.355.5+17.2
- LongBench v2 easy (w/o CoT)GPT-4o by 14.757.442.7+14.7
- LongBench v2 medium (w/o CoT)GPT-4o by 14.352.438.1+14.3
- MGSMGPT-4o by 14.390.576.2+14.3
- AIME 2024Qwen2.5 Instruct 72B by 149.323.3+14
- naturalquestions_closedbookGPT-4o by 13.749.635.9+13.7
- C-EvalQwen2.5 Instruct 72B by 13.27689.2+13.2
- Artificial Analysis Coding IndexGPT-4o by 12.324.211.9+12.3
- FRAMES (Acc.)GPT-4o by 10.780.569.8+10.7
- LongBench v2 short (w/ CoT) (Pass@1 with chain-of-thought)GPT-4o by 10.759.648.9+10.7
- Chinese SimpleQA (C-SimpleQA)GPT-4o by 10.358.748.4+10.3
- HarmBenchGPT-4o by 10.182.972.8+10.1
- LongBench v2 hard (w/ CoT) (Pass@1 with chain-of-thought)GPT-4o by 8.949.740.8+8.9
- DROP (F1)GPT-4o by 8.689.280.6+8.6
- LongBench v2 overall (w/o CoT)GPT-4o by 850.142.1+8
- LongBench v2 overall (w/ CoT) (Pass@1 with chain-of-thought)GPT-4o by 7.951.443.5+7.9
- LongBench v2 medium (w/ CoT) (Pass@1 with chain-of-thought)GPT-4o by 7.748.640.9+7.7
- LongBench v2 short (w/o CoT)GPT-4o by 7.753.345.6+7.7
- Aider-Edit (Acc.)GPT-4o by 7.572.965.4+7.5
- DROP (3-shot F1)GPT-4o by 783.776.7+7
- DROPGPT-4o by 6.783.476.7+6.7
- MMMLUGPT-4o by 6.681.474.8+6.6
- LongBench v2 easy (w/ CoT) (Pass@1 with chain-of-thought)GPT-4o by 6.354.247.9+6.3
- τ²-Bench Telecom (AA run)Qwen2.5 Instruct 72B by 5.628.934.5+5.6
- CLUEWSCGPT-4o by 5.487.982.5+5.4
- MATH-500 (EM)Qwen2.5 Instruct 72B by 5.474.680+5.4
- NarrativeQAGPT-4o by 579.574.5+5
- GSM8KQwen2.5 Instruct 72B by 4.990.995.8+4.9
- LongBench v2 long (w/o CoT)Qwen2.5 Instruct 72B by 4.240.244.4+4.2
- LongBench v2 hard (w/o CoT)GPT-4o by 3.845.641.8+3.8
- HumanEvalGPT-4o by 3.690.286.6+3.6
- MMLU-ProGPT-4o by 3.674.771.1+3.6
- AIR-Bench 2024GPT-4o by 3.462.459+3.4
- HumanEval-Mul (Pass@1)GPT-4o by 3.280.577.3+3.2
- MMLUQwen2.5 Instruct 72B by 3.273.877+3.2
- IFEvalQwen2.5 Instruct 72B by 3.18184.1+3.1
- IFEval (avg)Qwen2.5 Instruct 72B by 3.184.187.2+3.1
- MATHQwen2.5 Instruct 72B by 3.185.388.4+3.1
- Arena HardQwen2.5 Instruct 72B by 1.979.381.2+1.9
- simple_safety_testsQwen2.5 Instruct 72B by 1.598.5100+1.5
- MMLU-ReduxGPT-4o by 1.28886.8+1.2
- MBPP+ (EvalPlus-augmented)tie76.277tie
- MT-Benchtie8.79.3tie
- OpenBookQAtie96.896.2tie
- XSTesttie97.397.9tie
- anthropic_red_teamtie99.199.6tie
- AA Intelligencetie7.37.7tie
- bbqtie95.195.4tie
Questions people ask
Which is better, GPT-4o or Qwen2.5 Instruct 72B?
GPT-4o wins three of the five areas where both have results: coding, facts and long documents. Qwen2.5 Instruct 72B wins none, but costs 95% less. They are level on reasoning and following instructions.
Which is better for coding?
GPT-4o. It wins 3 of the 3 coding tests both models report; Qwen2.5 Instruct 72B wins none.
Which is cheaper?
GPT-4o costs $5.00 per million input tokens and $15.00 per million output tokens; Qwen2.5 Instruct 72B costs $0.47 and $0.49. That makes Qwen2.5 Instruct 72B about 95% cheaper for the same work.
How do you compare the two?
We use the 62 benchmark tests both models have published scores on. The verdict counts the 11 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 51 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.