Model comparison
GPT-4 vs Qwen2.5-Coder-32B
Qwen2.5-Coder-32B is the stronger model overall, scoring 33.4 to 29.1 on the Noometry Index.
Last verified . 20 shared benchmarks.
Alibaba (Qwen)Summary
- They share 20 benchmarks with published results for both. GPT-4 scores higher in 3 categories and Qwen2.5-Coder-32B in 5 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where Qwen2.5-Coder-32B leads 33.3 to 10.8.
- Qwen2.5-Coder-32B is cheaper at $0.66 / $1 per million input/output tokens, against $30 / $60 for GPT-4.
- Qwen2.5-Coder-32B accepts more context: 33K tokens versus 8K.
- Qwen2.5-Coder-32B has downloadable open weights; the other is API-only.
Side by side
| GPT-4 | Qwen2.5-Coder-32B | |
|---|---|---|
| Provider | OpenAI | Alibaba (Qwen) |
| Noometry Index | 29.1 | 33.4 |
| Released | 2023-03-14 | 2024-09-18 |
| Weights | Proprietary | Open |
| Context window | 8K | 33K |
| Max output | 8K | 29K |
| Input $ / M tokens | $30 | $0.66 |
| Output $ / M tokens | $60 | $1 |
| Results tracked | 38 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-4 leads
GPT-4: 31.6 (#283), Qwen2.5-Coder-32B: 22.6 (#333)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| BigCodeBench Instruct | 46% | 49% |
| LMArena Coding | 1254 | 1276 |
| BigCodeBench Complete | 57.2% | 58% |
| HumanEval+ | 79.3% | 87.2% |
| SWE-bench Verified (bash only) | — | 9% |
| Aider Polyglot | — | 16.4% |
| WeirdML | 12.4% | — |
| LiveBench Coding | — | 56.9% |
| MBPP+ | — | 77% |
Agentic & Tool Use Not comparable
GPT-4: —, Qwen2.5-Coder-32B: —
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| METR Time Horizons | 36.1% | — |
Reasoning Qwen2.5-Coder-32B leads
GPT-4: 17.8 (#289), Qwen2.5-Coder-32B: 21.2 (#225)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1251 |
| Epoch Capabilities Index | 125.89 | 119.49 |
| HellaSwag | 95.3% | 83% |
| WinoGrande | 87.5% | 80.8% |
| Chess Puzzles | 4% | — |
| LiveBench Reasoning | — | 42.1% |
| Mystery Game Puzzles | 12% | — |
| DTBench | 62.7% | — |
| LiveBench Data Analysis | — | 49.9% |
| LMCA | 17.1% | — |
| BIG-Bench Hard | 75.1% | — |
| ForecastBench | 57.8 | — |
| LiveBench | — | 46.2% |
Math Qwen2.5-Coder-32B leads
GPT-4: 10.8 (#309), Qwen2.5-Coder-32B: 33.3 (#204)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Math | 1269 | 1251 |
| GSM8K | 92% | 93% |
| OTIS Mock AIME 2024-2025 | 1.1% | — |
| LiveBench Math | — | 46.6% |
| MATH Level 5 | 23% | — |
Knowledge Qwen2.5-Coder-32B leads
GPT-4: 18.4 (#282), Qwen2.5-Coder-32B: 33.4 (#203)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Expert | 1211 | 1221 |
| MMLU | 86.4% | 79.1% |
| GPQA Diamond | 35.7% | — |
| ARC (AI2) Challenge | — | 70.5% |
| TriviaQA | 84.8% | — |
Multilingual GPT-4 leads
GPT-4: 40.6 (#215), Qwen2.5-Coder-32B: 37.8 (#235)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Non-English | 1246 | 1205 |
| LMArena Chinese | 1242 | 1222 |
| LMArena Russian | 1251 | 1228 |
| LMArena French | 1283 | — |
| LMArena German | 1251 | — |
| LMArena Japanese | 1209 | — |
| LMArena Korean | 1184 | — |
| LMArena Spanish | 1261 | — |
Instruction Following GPT-4 leads
GPT-4: 65.3 (#222), Qwen2.5-Coder-32B: 61.4 (#245)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Instruction Following | 1241 | 1223 |
| LiveBench Instruction Following | — | 58.7% |
Long Context Too close to call
GPT-4: 37.7 (#212), Qwen2.5-Coder-32B: 38.0 (#208)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Longer Query | 1244 | 1251 |
Writing & Preference Qwen2.5-Coder-32B leads
GPT-4: 34.9 (#268), Qwen2.5-Coder-32B: 41.6 (#240)
| Benchmark | GPT-4 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Text | 1263 | 1230 |
| LMArena Creative Writing | 1244 | 1174 |
| LMArena Multi-Turn | 1257 | 1222 |
| EQ-Bench Creative Writing | 752 | — |
| LiveBench Language | — | 23.3% |
Frequently asked questions
Is GPT-4 better than Qwen2.5-Coder-32B?
Qwen2.5-Coder-32B is the stronger model overall, scoring 33.4 to 29.1 on the Noometry Index.
Which is cheaper, GPT-4 or Qwen2.5-Coder-32B?
Qwen2.5-Coder-32B is cheaper. It lists at $0.66 per million input tokens and $1 per million output tokens; GPT-4 lists at $30 and $60.
Is GPT-4 or Qwen2.5-Coder-32B better for coding?
GPT-4 scores higher on coding benchmarks: 31.6 versus 22.6 in the Noometry coding category.
Which has the bigger context window?
Qwen2.5-Coder-32B does, with 33K tokens against 8K.
How many benchmarks do GPT-4 and Qwen2.5-Coder-32B share?
20 benchmarks have published results for both models. GPT-4 has 38 scored results on Noometry and Qwen2.5-Coder-32B has 31.