GPT-4o vs Qwen2.5 72B Instruct
Qwen2.5 72B Instruct costs 25.0x less per output token than GPT-4o ($0.40 vs $10.00 per million). Updated October 21, 2026; 12 benchmarks have a result for both.
GPT-4o or Qwen2.5 72B Instruct: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick GPT-4o for a larger context window (128K against 33K tokens) and work measured by τ²-Bench Telecom or Humanity's Last Exam (Artificial Analysis run), where it leads.
- Shared results
- 12
- GPT-4o only
- 15
- Qwen2.5 72B Instruct only
- 41
- Like-for-like categories
- 6 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
No call
Only Qwen2.5 72B Instruct has a AA Intelligence Index result, 7.7.
Directional only
Web design
Design Arena: websites
No call
Only GPT-4o has a Design Arena: websites result, 831.
Directional only
Long documents
Listed context window
GPT-4o
GPT-4o takes 128K tokens against 33K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen2.5 72B Instruct
GPT-4o $0.0075, Qwen2.5 72B Instruct $0.0006 per turn.
List prices
At a glance
- Coding index
- GPT-4o not benchmarked, Qwen2.5 72B Instruct not benchmarked.
- Widest category
- Agentic work, 9.4 points apart on average over 1 shared benchmark.
- Widest single gap
- τ²-Bench Telecom: GPT-4o 25.1% against Qwen2.5 72B Instruct 34.5%, 9.4 points apart on a 0 to 100 scale.
- Output price
- GPT-4o $10.00 against Qwen2.5 72B Instruct $0.40 per million output tokens, 25.0x apart.
- Context window
- GPT-4o takes 128K tokens, Qwen2.5 72B Instruct 33K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | GPT-4o | Bars | Qwen2.5 72B Instruct | Ties |
|---|---|---|---|---|
| Coding | 2 | 0 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 1 | 1 | ||
| Math | 0 | 2 | ||
| Knowledge | 2 | 0 | ||
| General | 2 | 1 | ||
| All shared boards | 7 | 5 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index20
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| SWE-bench Verified | 38.8% | Not measured | Only GPT-4o measured |
| SWE-bench Lite | 38.0% | Not measured | Only GPT-4o measured |
| SWE-bench (full test set) | 18.8% | Not measured | Only GPT-4o measured |
| SWE-bench Multimodal | 30.4% | Not measured | Only GPT-4o measured |
| LiveCodeBench (Artificial Analysis run) | 30.9% | 27.6% | GPT-4o |
| Terminal-Bench Hard (Artificial Analysis run) | 8.3% | 4.5% | GPT-4o |
| WeirdML | Not measured | 16.0% | Only Qwen2.5 72B Instruct measured |
| LMArena Text: coding | Not measured | 1292 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: software and IT | Not measured | 1292 | Only Qwen2.5 72B Instruct measured |
| SciCode | 1.5% | Not measured | Only GPT-4o measured |
| Neither model has a result on the other 29 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 7.7 | Only Qwen2.5 72B Instruct measured |
| Epoch Capabilities Index | Not measured | 129.0 | Only Qwen2.5 72B Instruct measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 25.1% | 34.5% | Qwen2.5 72B Instruct |
| METR time horizons | Not measured | 35.8% | Only Qwen2.5 72B Instruct measured |
| GDPval | 9.9% | Not measured | Only GPT-4o measured |
| BALROG | Not measured | 16.2% | Only Qwen2.5 72B Instruct measured |
| TheAgentCompany | Not measured | 5.7% | Only Qwen2.5 72B Instruct measured |
| Neither model has a result on the other 36 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 54.3% | 49.1% | GPT-4o |
| Humanity's Last Exam (Artificial Analysis run) | 2.4% | 3.6% | Qwen2.5 72B Instruct |
| GPQA Diamond (Epoch AI run) | Not measured | 49.1% | Only Qwen2.5 72B Instruct measured |
| ARC-AGI | 4.5% | Not measured | Only GPT-4o measured |
| ARC-AGI-2 | 0.0% | Not measured | Only GPT-4o measured |
| BIG-Bench Hard | Not measured | 79.8% | Only Qwen2.5 72B Instruct measured |
| LMArena Text: hard prompts | Not measured | 1271 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: expert | Not measured | 1245 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: hard prompts in English | Not measured | 1279 | Only Qwen2.5 72B Instruct measured |
| Neither model has a result on the other 13 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| AA Math Index | 6.0 | 14.0 | Qwen2.5 72B Instruct |
| AIME 2025 (Artificial Analysis run) | 6.0% | 14.0% | Qwen2.5 72B Instruct |
| OTIS Mock AIME | Not measured | 8.1% | Only Qwen2.5 72B Instruct measured |
| MATH level 5 | Not measured | 63.2% | Only Qwen2.5 72B Instruct measured |
| LMArena Text: math | Not measured | 1283 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: mathematical work | Not measured | 1290 | Only Qwen2.5 72B Instruct measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | 74.8% | 72.0% | GPT-4o |
| MMLU | 88.1% | 85.3% | GPT-4o |
| AI2 Reasoning Challenge | Not measured | 94.5% | Only Qwen2.5 72B Instruct measured |
| LMArena Text: science | Not measured | 1275 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: medicine and healthcare | Not measured | 1258 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: legal and government | Not measured | 1288 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: business and finance | Not measured | 1253 | Only Qwen2.5 72B Instruct measured |
| Neither model has a result on the other 11 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| Design Arena: websites | 831 | Not measured | Only GPT-4o measured |
| Design Arena: UI components | 885 | Not measured | Only GPT-4o measured |
| Design Arena: game dev | 906 | Not measured | Only GPT-4o measured |
| Design Arena: data viz | 858 | Not measured | Only GPT-4o measured |
| Design Arena: 3D | 869 | Not measured | Only GPT-4o measured |
| Design Arena: all code categories | 861 | Not measured | Only GPT-4o measured |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index21
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| IFBench | 34.3% | 36.9% | Qwen2.5 72B Instruct |
| Long context reasoning (AA-LCR) | 49.3% | Not measured | Only GPT-4o measured |
| DTBench | 64.5% | 62.9% | GPT-4o |
| LMCA | 16.6% | 13.4% | GPT-4o |
| HellaSwag | Not measured | 84.8% | Only Qwen2.5 72B Instruct measured |
| WinoGrande | Not measured | 82.3% | Only Qwen2.5 72B Instruct measured |
| PIQA | Not measured | 82.6% | Only Qwen2.5 72B Instruct measured |
| LMArena Text | Not measured | 1269 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: instruction following | Not measured | 1254 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: creative writing | Not measured | 1221 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: English | Not measured | 1283 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: non-English | Not measured | 1252 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: Chinese | Not measured | 1273 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: German | Not measured | 1234 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: Japanese | Not measured | 1180 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: Korean | Not measured | 1188 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: French | Not measured | 1280 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: Spanish | Not measured | 1255 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: Russian | Not measured | 1264 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: multi-turn | Not measured | 1272 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: longer queries | Not measured | 1281 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: writing and language | Not measured | 1247 | Only Qwen2.5 72B Instruct measured |
| LMArena Text: entertainment and media | Not measured | 1211 | Only Qwen2.5 72B Instruct measured |
| Neither model has a result on the other 23 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-4o | Qwen2.5 72B Instruct | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Qwen2.5 72B Instruct costs less
Repository review
50K input + 3K output tokens
Qwen2.5 72B Instruct costs less
Agent loop
200K input + 10K output tokens
Qwen2.5 72B Instruct costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | OpenAI GPT-4o | Qwen2.5 72B Instruct |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | Open weights |
| Released | May 13, 2024 | September 19, 2024 |
| Context window | 128K | 33K |
| Input $/Mtok | $2.50 | $0.36 |
| Output $/Mtok | $10.00 | $0.40 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-4o better than Qwen2.5 72B Instruct for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick GPT-4o for a larger context window (128K against 33K tokens) and work measured by τ²-Bench Telecom or Humanity's Last Exam (Artificial Analysis run), where it leads.
Which is cheaper, GPT-4o or Qwen2.5 72B Instruct?
GPT-4o costs $2.50 per million input tokens and $10.00 per million output tokens; Qwen2.5 72B Instruct costs $0.36 and $0.40. Cost per solved task is known for neither of them.
Which has the bigger context window, GPT-4o or Qwen2.5 72B Instruct?
GPT-4o, with 128K tokens against 33K.
How many benchmarks is the GPT-4o vs Qwen2.5 72B Instruct comparison based on?
12 benchmarks have a result for both models as of October 21, 2026. GPT-4o's coding index rests on 0 independent benchmarks and Qwen2.5 72B Instruct's on 0. Where only one model has a result, the page says so instead of naming a winner.