# GPT-5.4 vs Qwen2.5 72B Instruct

> GPT-5.4 is the stronger model overall, scoring 59.5 to 31.5 on the Noometry Index. Qwen2.5 72B Instruct costs 2.3× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/gpt-5-4-vs-qwen2-5-72b-instruct
- Last updated: 2026-10-10
- Shared benchmarks: 25

## Summary

- They share 25 benchmarks with published results for both. GPT-5.4 scores higher in 9 categories and Qwen2.5 72B Instruct in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.4 leads 73.3 to 18.1.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 97.8% for GPT-5.4 and 8.1% for Qwen2.5 72B Instruct.
- Qwen2.5 72B Instruct is cheaper at $1.40 / $5.60 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 131K.
- Qwen2.5 72B Instruct has downloadable open weights; the other is API-only.

## Snapshot

| | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| Provider | OpenAI | Alibaba (Qwen) |
| Noometry Index | 59.5 | 31.5 |
| Rank | 14 | 259 |
| Context | 1.05M | 131K |
| Input $/M | $2.50 | $1.40 |
| Output $/M | $15 | $5.60 |
| Weights | Proprietary | Open |

## Coding

- GPT-5.4: 52.8 (#28)
- Qwen2.5 72B Instruct: 32.8 (#256)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| WeirdML | 77.7% | 16% |
| LMArena Coding | 1497 | 1293 |
| SWE-bench Verified | 76.9% | — |
| DeepSWE | 51.8% | — |
| LMArena WebDev | 1465 | — |
| SciCode | 56.6% | — |
| GSO | 31.4% | — |
| BigCodeBench Instruct | — | 45.8% |
| MirrorCode | 15.6% | — |
| BigCodeBench Complete | — | 55.9% |
| ALE-Bench | 1,607 | — |
| AlgoTune | 1.85 | — |

## Agentic & Tool Use

- GPT-5.4: 46.5 (#12)
- Qwen2.5 72B Instruct: 22.0 (#124)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| METR Time Horizons | 74.3% | 35.8% |
| Terminal-Bench | 81.8% | — |
| APEX-Agents | 52.4% | — |
| TheAgentCompany | — | 5.7% |
| τ²-bench Banking | 39.4% | — |
| DeepResearch Bench | 35.1% | — |
| PostTrainBench | 19% | — |
| BALROG | — | 16.2% |
| GBAEval | 45.1% | — |
| LMArena Search | 1197 | — |
| Vending-Bench 2 | 6,144 | — |

## Reasoning

- GPT-5.4: 61.7 (#17)
- Qwen2.5 72B Instruct: 21.4 (#197)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Hard Prompts | 1485 | 1271 |
| DTBench | 94.4% | 62.9% |
| LMCA | 52% | 13.4% |
| Epoch Capabilities Index | 156.92 | 129.01 |
| ForecastBench | 59.5 | 57.5 |
| ARC-AGI-2 | 74% | — |
| Kagi LLM Benchmark | 63.8% | — |
| NYT Connections (extended) | 91.3% | — |
| ARC-AGI-1 | 93.7% | — |
| CritPt | 23.4% | — |
| Chess Puzzles | 44% | — |
| EnigmaEval | 16% | — |
| Thematic Generalization | 80% | — |
| EBR-Bench | 25.4% | — |
| Mystery Game Puzzles | 37% | — |
| BIG-Bench Hard | — | 79.8% |
| HellaSwag | — | 84.8% |
| PIQA | — | 82.6% |
| WinoGrande | — | 82.3% |

## Math

- GPT-5.4: 73.3 (#14)
- Qwen2.5 72B Instruct: 18.1 (#281)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 97.8% | 8.1% |
| LMArena Math | 1488 | 1283 |
| FrontierMath (Tiers 1-3) | 78.6% | — |
| FrontierMath Tier 4 | 49% | — |
| MathArena Final-Answer Competitions | 83.1% | — |
| ProofBench | 56% | — |
| Omni-MATH | — | 33% |
| MATH Level 5 | — | 63.2% |
| FrontierMath (Feb 2025 set) | 47.6% | — |
| FrontierMath Tier 4 (v1) | 27.1% | — |

## Knowledge

- GPT-5.4: 65.3 (#13)
- Qwen2.5 72B Instruct: 26.9 (#247)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| GPQA Diamond | 93.3% | 49.1% |
| LMArena Expert | 1507 | 1245 |
| Humanity's Last Exam | 36.2% | — |
| SimpleQA Verified | 45.1% | — |
| MMLU-Pro | — | 63.1% |
| Confabulations | — | 19.1% |
| Vectara Hallucination Rate | 7% | — |
| GPQA (HELM) | — | 42.6% |
| ARC (AI2) Challenge | — | 94.5% |
| MMLU | — | 85.3% |
| TriviaQA | — | 71.9% |

## Multimodal

- GPT-5.4: 43.8 (#19)
- Qwen2.5 72B Instruct: —

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Vision | 1302 | — |
| Blueprint-Bench 2 | 27.1% | — |
| Furniture Assembly | 37.5% | — |
| LMArena Document | 1471 | — |

## Multilingual

- GPT-5.4: 56.4 (#22)
- Qwen2.5 72B Instruct: 41.2 (#208)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Non-English | 1465 | 1252 |
| LMArena Chinese | 1520 | 1272 |
| LMArena French | 1493 | 1279 |
| LMArena German | 1472 | 1234 |
| LMArena Japanese | 1484 | 1180 |
| LMArena Korean | 1449 | 1188 |
| LMArena Russian | 1480 | 1264 |
| LMArena Spanish | 1453 | 1255 |

## Instruction Following

- GPT-5.4: 77.3 (#25)
- Qwen2.5 72B Instruct: 65.7 (#216)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Instruction Following | 1469 | 1254 |
| IFEval | — | 80.6% |

## Long Context

- GPT-5.4: 50.3 (#8)
- Qwen2.5 72B Instruct: 38.9 (#183)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Longer Query | 1474 | 1282 |
| CL-bench | 27.9% | — |
| CL-bench Life | 21.7% | — |

## Writing & Preference

- GPT-5.4: 72.1 (#18)
- Qwen2.5 72B Instruct: 46.9 (#210)

| Benchmark | GPT-5.4 | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Text | 1469 | 1269 |
| LMArena Creative Writing | 1439 | 1221 |
| LMArena Multi-Turn | 1482 | 1272 |
| EQ-Bench Creative Writing | 1840 | — |
| WildBench | — | 80.2% |
| EQ-Bench 4 | 1272 | — |

## FAQ

### Is GPT-5.4 better than Qwen2.5 72B Instruct?

GPT-5.4 is the stronger model overall, scoring 59.5 to 31.5 on the Noometry Index. Qwen2.5 72B Instruct costs 2.3× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.

### Which is cheaper, GPT-5.4 or Qwen2.5 72B Instruct?

Qwen2.5 72B Instruct is cheaper. It lists at $1.40 per million input tokens and $5.60 per million output tokens; GPT-5.4 lists at $2.50 and $15.

### Is GPT-5.4 or Qwen2.5 72B Instruct better for coding?

GPT-5.4 scores higher on coding benchmarks: 52.8 versus 32.8 in the Noometry coding category.

### Which has the bigger context window?

GPT-5.4 does, with 1.05M tokens against 131K.

### How many benchmarks do GPT-5.4 and Qwen2.5 72B Instruct share?

25 benchmarks have published results for both models. GPT-5.4 has 68 scored results on Noometry and Qwen2.5 72B Instruct has 43.