Model comparison
Llama 3.1-70B vs Qwen2.5-Coder-32B
Qwen2.5-Coder-32B is the stronger model overall, scoring 32.9 to 29.4 on the Noometry Index. Llama 3.1-70B costs 1.9× less per token, which makes it the better buy when Qwen2.5-Coder-32B's lead doesn't matter for your workload.
Last verified . 16 shared benchmarks.
Meta
Alibaba (Qwen)Summary
- They share 16 benchmarks with published results for both. Llama 3.1-70B scores higher in 4 categories and Qwen2.5-Coder-32B in 4 categories; 5 gaps are clear of the uncertainty.
- The widest gap is in math, where Qwen2.5-Coder-32B leads 30.8 to 12.7.
- Llama 3.1-70B is cheaper at $0.40 / $0.40 per million input/output tokens, against $0.66 / $1 for Qwen2.5-Coder-32B.
- Llama 3.1-70B accepts more context: 128K tokens versus 33K.
Side by side
| Llama 3.1-70B | Qwen2.5-Coder-32B | |
|---|---|---|
| Provider | Meta | Alibaba (Qwen) |
| Noometry Index | 29.4 | 32.9 |
| Released | 2024-07-23 | 2024-09-18 |
| Weights | Open | Open |
| Context window | 128K | 33K |
| Max output | 4K | 29K |
| Input $ / M tokens | $0.40 | $0.66 |
| Output $ / M tokens | $0.40 | $1 |
| Results tracked | 35 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Llama 3.1-70B leads
Llama 3.1-70B: 29.8 (#292), Qwen2.5-Coder-32B: 22.4 (#328)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| BigCodeBench Instruct | 46.1% | 49% |
| LMArena Coding | 1260 | 1276 |
| BigCodeBench Complete | 54.8% | 58% |
| SWE-bench Verified (bash only) | — | 9% |
| Aider Polyglot | — | 16.4% |
| WeirdML | 9% | — |
| LiveBench Coding | — | 56.9% |
| HumanEval+ | — | 87.2% |
| MBPP+ | — | 77% |
Agentic & Tool Use Not comparable
Llama 3.1-70B: 24.9 (#104), Qwen2.5-Coder-32B: —
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| TheAgentCompany | 6.9% | — |
| BALROG | 27.9% | — |
Reasoning Too close to call
Llama 3.1-70B: 20.8 (#215), Qwen2.5-Coder-32B: 20.4 (#220)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1251 |
| Epoch Capabilities Index | 125.91 | 119.41 |
| LiveBench Reasoning | — | 42.1% |
| DTBench | 60% | — |
| LiveBench Data Analysis | — | 49.9% |
| LMCA | 14.8% | — |
| HellaSwag | — | 83% |
| LiveBench | — | 46.2% |
| WinoGrande | — | 80.8% |
Math Qwen2.5-Coder-32B leads
Llama 3.1-70B: 12.7 (#298), Qwen2.5-Coder-32B: 30.8 (#203)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Math | 1252 | 1251 |
| OTIS Mock AIME 2024-2025 | 3.6% | — |
| Omni-MATH | 21% | — |
| LiveBench Math | — | 46.6% |
| MATH Level 5 | 36.7% | — |
| GSM8K | — | 93% |
Knowledge Qwen2.5-Coder-32B leads
Llama 3.1-70B: 24.1 (#264), Qwen2.5-Coder-32B: 32.9 (#200)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Expert | 1209 | 1221 |
| MMLU | 80.1% | 79.1% |
| GPQA Diamond | 44.2% | — |
| MMLU-Pro | 65.3% | — |
| GPQA (HELM) | 42.6% | — |
| ARC (AI2) Challenge | — | 70.5% |
Multilingual Too close to call
Llama 3.1-70B: 38.9 (#220), Qwen2.5-Coder-32B: 38.0 (#230)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Non-English | 1219 | 1205 |
| LMArena Chinese | 1215 | 1222 |
| LMArena Russian | 1234 | 1228 |
| LMArena French | 1260 | — |
| LMArena German | 1222 | — |
| LMArena Japanese | 1132 | — |
| LMArena Korean | 1140 | — |
| LMArena Spanish | 1253 | — |
Instruction Following Llama 3.1-70B leads
Llama 3.1-70B: 65.5 (#218), Qwen2.5-Coder-32B: 61.6 (#240)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Instruction Following | 1231 | 1223 |
| LiveBench Instruction Following | — | 58.7% |
| IFEval | 82.1% | — |
Long Context Too close to call
Llama 3.1-70B: 37.6 (#208), Qwen2.5-Coder-32B: 37.9 (#204)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Longer Query | 1241 | 1251 |
Writing & Preference Qwen2.5-Coder-32B leads
Llama 3.1-70B: 35.6 (#262), Qwen2.5-Coder-32B: 41.8 (#235)
| Benchmark | Llama 3.1-70B | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Text | 1261 | 1230 |
| LMArena Creative Writing | 1232 | 1174 |
| LMArena Multi-Turn | 1256 | 1222 |
| EQ-Bench Creative Writing | 784 | — |
| WildBench | 75.8% | — |
| LiveBench Language | — | 23.3% |
Frequently asked questions
Is Llama 3.1-70B better than Qwen2.5-Coder-32B?
Qwen2.5-Coder-32B is the stronger model overall, scoring 32.9 to 29.4 on the Noometry Index. Llama 3.1-70B costs 1.9× 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 3.1-70B or Qwen2.5-Coder-32B?
Llama 3.1-70B is cheaper. It lists at $0.40 per million input tokens and $0.40 per million output tokens; Qwen2.5-Coder-32B lists at $0.66 and $1.
Is Llama 3.1-70B or Qwen2.5-Coder-32B better for coding?
Llama 3.1-70B scores higher on coding benchmarks: 29.8 versus 22.4 in the Noometry coding category.
Which has the bigger context window?
Llama 3.1-70B does, with 128K tokens against 33K.
How many benchmarks do Llama 3.1-70B and Qwen2.5-Coder-32B share?
16 benchmarks have published results for both models. Llama 3.1-70B has 35 scored results on Noometry and Qwen2.5-Coder-32B has 31.