Model comparison
Qwen2.5 72B Instruct vs Qwen3-30B-A3B
Qwen3-30B-A3B is the stronger model overall, scoring 38.9 to 31.9 on the Noometry Index.
Last verified . 24 shared benchmarks.
Alibaba (Qwen)Summary
- They share 24 benchmarks with published results for both. Qwen2.5 72B Instruct scores higher in 2 categories and Qwen3-30B-A3B in 7 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where Qwen3-30B-A3B leads 37.4 to 19.3.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 8.1% for Qwen2.5 72B Instruct and 70.3% for Qwen3-30B-A3B.
- Qwen3-30B-A3B is cheaper at $0.12 / $0.50 per million input/output tokens, against $1.40 / $5.60 for Qwen2.5 72B Instruct.
- Qwen2.5 72B Instruct accepts more context: 131K tokens versus 41K.
Side by side
| Qwen2.5 72B Instruct | Qwen3-30B-A3B | |
|---|---|---|
| Provider | Alibaba (Qwen) | Alibaba (Qwen) |
| Noometry Index | 31.9 | 38.9 |
| Released | 2024-09 | 2025-04-28 |
| Weights | Open | Open |
| Context window | 131K | 41K |
| Max output | 8K | 16K |
| Input $ / M tokens | $1.40 | $0.12 |
| Output $ / M tokens | $5.60 | $0.50 |
| Results tracked | 43 | 32 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 33.2 (#260), Qwen3-30B-A3B: 37.5 (#194)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| WeirdML | 16% | 29.8% |
| LMArena Coding | 1292 | 1416 |
| SciCode | — | 33.3% |
| BigCodeBench Instruct | 45.8% | — |
| BigCodeBench Complete | 55.9% | — |
Agentic & Tool Use Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 22.1 (#133), Qwen3-30B-A3B: 29.8 (#82)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 41.4% |
| TheAgentCompany | 5.7% | — |
| BALROG | 16.2% | — |
| METR Time Horizons | 35.8% | — |
Reasoning Too close to call
Qwen2.5 72B Instruct: 22.3 (#199), Qwen3-30B-A3B: 22.2 (#204)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| LMArena Hard Prompts | 1271 | 1398 |
| DTBench | 62.9% | 69.3% |
| LMCA | 13.4% | 22.4% |
| Epoch Capabilities Index | 129 | 139.63 |
| Kagi LLM Benchmark | — | 54.9% |
| CritPt | — | 0.3% |
| Chess Puzzles | — | 8% |
| BIG-Bench Hard | 79.8% | — |
| ForecastBench | 57.5 | — |
| HellaSwag | 84.8% | — |
| PIQA | 82.6% | — |
| WinoGrande | 82.3% | — |
Math Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 19.3 (#287), Qwen3-30B-A3B: 37.4 (#157)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 8.1% | 70.3% |
| LMArena Math | 1283 | 1394 |
| MathArena Final-Answer Competitions | — | 47.8% |
| Omni-MATH | 33% | — |
| MATH Level 5 | 63.2% | — |
Knowledge Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 27.0 (#253), Qwen3-30B-A3B: 41.8 (#105)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| GPQA Diamond | 49.1% | 70.1% |
| Confabulations | 19.1% | 12.3% |
| LMArena Expert | 1245 | 1396 |
| MMLU-Pro | 63.1% | — |
| GPQA (HELM) | 42.6% | — |
| ARC (AI2) Challenge | 94.5% | — |
| MMLU | 85.3% | — |
| TriviaQA | 71.9% | — |
Multilingual Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 41.0 (#213), Qwen3-30B-A3B: 49.5 (#132)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| LMArena Non-English | 1252 | 1372 |
| LMArena Chinese | 1272 | 1433 |
| LMArena French | 1280 | 1418 |
| LMArena German | 1234 | 1380 |
| LMArena Japanese | 1180 | 1337 |
| LMArena Korean | 1188 | 1331 |
| LMArena Russian | 1264 | 1370 |
| LMArena Spanish | 1256 | 1404 |
Instruction Following Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 65.5 (#221), Qwen3-30B-A3B: 72.0 (#142)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| LMArena Instruction Following | 1254 | 1363 |
| IFEval | 80.6% | — |
Long Context Qwen2.5 72B Instruct leads
Qwen2.5 72B Instruct: 38.9 (#188), Qwen3-30B-A3B: 31.0 (#283)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| LMArena Longer Query | 1282 | 1379 |
| Fiction.LiveBench | — | 40.6% |
Writing & Preference Qwen3-30B-A3B leads
Qwen2.5 72B Instruct: 46.7 (#215), Qwen3-30B-A3B: 55.6 (#143)
| Benchmark | Qwen2.5 72B Instruct | Qwen3-30B-A3B |
|---|---|---|
| LMArena Text | 1269 | 1384 |
| LMArena Creative Writing | 1221 | 1317 |
| LMArena Multi-Turn | 1272 | 1378 |
| Short-Story Creative Writing | — | 75.3% |
| WildBench | 80.2% | — |
Frequently asked questions
Is Qwen2.5 72B Instruct better than Qwen3-30B-A3B?
Qwen3-30B-A3B is the stronger model overall, scoring 38.9 to 31.9 on the Noometry Index.
Which is cheaper, Qwen2.5 72B Instruct or Qwen3-30B-A3B?
Qwen3-30B-A3B is cheaper. It lists at $0.12 per million input tokens and $0.50 per million output tokens; Qwen2.5 72B Instruct lists at $1.40 and $5.60.
Is Qwen2.5 72B Instruct or Qwen3-30B-A3B better for coding?
Qwen3-30B-A3B scores higher on coding benchmarks: 37.5 versus 33.2 in the Noometry coding category.
Which has the bigger context window?
Qwen2.5 72B Instruct does, with 131K tokens against 41K.
How many benchmarks do Qwen2.5 72B Instruct and Qwen3-30B-A3B share?
24 benchmarks have published results for both models. Qwen2.5 72B Instruct has 43 scored results on Noometry and Qwen3-30B-A3B has 32.