Model comparison
Llama 4 Maverick vs Qwen2.5-Coder-32B
Qwen2.5-Coder-32B is the stronger model overall, scoring 32.9 to 30.6 on the Noometry Index. Llama 4 Maverick costs 2.5× less per token, which makes it the better buy when Qwen2.5-Coder-32B's lead doesn't matter for your workload.
Last verified . 17 shared benchmarks.
Meta
Alibaba (Qwen)Summary
- They share 17 benchmarks with published results for both. Llama 4 Maverick scores higher in 4 categories and Qwen2.5-Coder-32B in 4 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Qwen2.5-Coder-32B leads 20.4 to 10.1.
- The biggest single-benchmark swing is SWE-bench Verified (bash only): 21% for Llama 4 Maverick and 9% for Qwen2.5-Coder-32B.
- Llama 4 Maverick is cheaper at $0.19 / $0.65 per million input/output tokens, against $0.66 / $1 for Qwen2.5-Coder-32B.
- Llama 4 Maverick accepts more context: 128K tokens versus 33K.
Side by side
| Llama 4 Maverick | Qwen2.5-Coder-32B | |
|---|---|---|
| Provider | Meta | Alibaba (Qwen) |
| Noometry Index | 30.6 | 32.9 |
| Released | 2025-04-05 | 2024-09-18 |
| Weights | Open | Open |
| Context window | 128K | 33K |
| Max output | 4K | 29K |
| Input $ / M tokens | $0.19 | $0.66 |
| Output $ / M tokens | $0.65 | $1 |
| Results tracked | 54 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Llama 4 Maverick leads
Llama 4 Maverick: 26.3 (#318), Qwen2.5-Coder-32B: 22.4 (#328)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| SWE-bench Verified (bash only) | 21% | 9% |
| Aider Polyglot | 15.6% | 16.4% |
| BigCodeBench Instruct | 49.7% | 49% |
| LMArena Coding | 1302 | 1276 |
| BigCodeBench Complete | 61.4% | 58% |
| SciCode | 33.1% | — |
| WeirdML | 24.5% | — |
| LiveBench Coding | — | 56.9% |
| ALE-Bench | 172.97 | — |
| HumanEval+ | — | 87.2% |
| MBPP+ | — | 77% |
Agentic & Tool Use Not comparable
Llama 4 Maverick: 27.9 (#86), Qwen2.5-Coder-32B: —
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| Berkeley Function Calling Leaderboard | 37.3% | — |
Reasoning Qwen2.5-Coder-32B leads
Llama 4 Maverick: 10.1 (#337), Qwen2.5-Coder-32B: 20.4 (#220)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Hard Prompts | 1281 | 1251 |
| Epoch Capabilities Index | 132.2 | 119.41 |
| ARC-AGI-2 | 0% | — |
| SimpleBench | 27.7% | — |
| Kagi LLM Benchmark | 55.9% | — |
| NYT Connections (extended) | 8% | — |
| ARC-AGI-1 | 4.4% | — |
| CritPt | 0% | — |
| EnigmaEval | 0.6% | — |
| LiveBench Reasoning | — | 42.1% |
| DTBench | 61.9% | — |
| LiveBench Data Analysis | — | 49.9% |
| LMCA | 15.9% | — |
| ForecastBench | 57.5 | — |
| HellaSwag | — | 83% |
| LiveBench | — | 46.2% |
| WinoGrande | — | 80.8% |
Math Qwen2.5-Coder-32B leads
Llama 4 Maverick: 24.6 (#256), Qwen2.5-Coder-32B: 30.8 (#203)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Math | 1299 | 1251 |
| OTIS Mock AIME 2024-2025 | 20.6% | — |
| Omni-MATH | 42.2% | — |
| LiveBench Math | — | 46.6% |
| MATH Level 5 | 73% | — |
| FrontierMath (Feb 2025 set) | 0.7% | — |
| GSM8K | — | 93% |
Knowledge Too close to call
Llama 4 Maverick: 33.3 (#196), Qwen2.5-Coder-32B: 32.9 (#200)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Expert | 1259 | 1221 |
| GPQA Diamond | 67% | — |
| Humanity's Last Exam | 5.7% | — |
| MMLU-Pro | 81% | — |
| Confabulations | 22.6% | — |
| Vectara Hallucination Rate | 8.2% | — |
| GPQA (HELM) | 65% | — |
| ARC (AI2) Challenge | — | 70.5% |
| MMLU | — | 79.1% |
Multimodal Not comparable
Llama 4 Maverick: 30.9 (#100), Qwen2.5-Coder-32B: —
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Vision | 1141 | — |
| GeoBench | 52% | — |
| SpatialViz-Bench | 31.8% | — |
Multilingual Llama 4 Maverick leads
Llama 4 Maverick: 42.4 (#190), Qwen2.5-Coder-32B: 38.0 (#230)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Non-English | 1269 | 1205 |
| LMArena Chinese | 1277 | 1222 |
| LMArena Russian | 1286 | 1228 |
| LMArena French | 1258 | — |
| LMArena German | 1291 | — |
| LMArena Japanese | 1207 | — |
| LMArena Korean | 1203 | — |
| LMArena Spanish | 1293 | — |
Instruction Following Llama 4 Maverick leads
Llama 4 Maverick: 71.9 (#140), Qwen2.5-Coder-32B: 61.6 (#240)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Instruction Following | 1267 | 1223 |
| LiveBench Instruction Following | — | 58.7% |
| IFEval | 90.8% | — |
Long Context Qwen2.5-Coder-32B leads
Llama 4 Maverick: 31.4 (#274), Qwen2.5-Coder-32B: 37.9 (#204)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Longer Query | 1280 | 1251 |
| Fiction.LiveBench | 46.2% | — |
Writing & Preference Qwen2.5-Coder-32B leads
Llama 4 Maverick: 39.0 (#247), Qwen2.5-Coder-32B: 41.8 (#235)
| Benchmark | Llama 4 Maverick | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Text | 1287 | 1230 |
| LMArena Creative Writing | 1266 | 1174 |
| LMArena Multi-Turn | 1289 | 1222 |
| Short-Story Creative Writing | 62% | — |
| EQ-Bench Creative Writing | 860 | — |
| WildBench | 80% | — |
| LiveBench Language | — | 23.3% |
Frequently asked questions
Is Llama 4 Maverick better than Qwen2.5-Coder-32B?
Qwen2.5-Coder-32B is the stronger model overall, scoring 32.9 to 30.6 on the Noometry Index. Llama 4 Maverick costs 2.5× less per token, which makes it the better buy when Qwen2.5-Coder-32B's lead doesn't matter for your workload.
Which is cheaper, Llama 4 Maverick or Qwen2.5-Coder-32B?
Llama 4 Maverick is cheaper. It lists at $0.19 per million input tokens and $0.65 per million output tokens; Qwen2.5-Coder-32B lists at $0.66 and $1.
Is Llama 4 Maverick or Qwen2.5-Coder-32B better for coding?
Llama 4 Maverick scores higher on coding benchmarks: 26.3 versus 22.4 in the Noometry coding category.
Which has the bigger context window?
Llama 4 Maverick does, with 128K tokens against 33K.
How many benchmarks do Llama 4 Maverick and Qwen2.5-Coder-32B share?
17 benchmarks have published results for both models. Llama 4 Maverick has 54 scored results on Noometry and Qwen2.5-Coder-32B has 31.