Compare / Qwen3.5 397B A17B vs MiniMax M3
Qwen3.5 397B A17B vs MiniMax M3
Which is cheaper, which is more capable, and which does better at each kind of work, from 22 results on 3 sources that measured both models.
Qwen3.5 397B A17B or MiniMax M3? The short answer
- MiniMax M3 is cheaper: $0.52 against $1.29 per million tokens, 2.5 times less.
- MiniMax M3 scores higher on the Artificial Analysis Intelligence Index: 29.2 against 21.4.
- On our business benchmarks, Qwen3.5 397B A17B ranks higher for agents and tool use and sticking to the facts.
- On our business benchmarks, MiniMax M3 ranks higher for professional work, reasoning and knowledge and writing.
At a glance
| Qwen3.5 397B A17B | MiniMax M3 | |
|---|---|---|
| Price per million tokens | $0.55 in · $3.50 out | $0.30 in · $1.20 out |
| Blended (3 in : 1 out) | $1.29 | $0.52 |
| Cheapest host, blended | $0.88 (Alibaba) | $0.41 (CoreWeave, FP4) |
| Intelligence Index | 21.4 | 29.2 |
| Spring Prompt overall | – | – |
| Output speed, tokens a second | 90 | 81 |
| Context window | 262,144 tokens | 1,048,576 tokens |
| Weights | Open | Open |
| Developer based in | China | China |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among models available today, for each kind of work, from our own benchmarks and licensed sources. The better rank is in green.
| Use case | Qwen3.5 397B A17B | MiniMax M3 |
|---|---|---|
| Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls | 65 of 161 | 71 of 161 |
| Professional workReal tasks from banking, consulting and law, business documents and freelance projects | 48 of 51 | 39 of 51 |
| Reasoning and knowledgeHard questions across science, maths and general knowledge | 69 of 195 | 36 of 195 |
| WritingWhat people prefer in blind comparisons, and judged writing quality | 66 of 187 | 61 of 187 |
| Sticking to the factsSummarising without inventing things, and factual answers | 72 of 153 | 94 of 153 |
Every shared result
| Benchmark | Metric | Qwen3.5 397B A17B | MiniMax M3 |
|---|---|---|---|
| APEX-Agentsreported | Tasks passed% of tasks | 24.9% | 37.7% |
| Arena (formerly LMArena)reported | OverallArena rating | 1,445 | 1,434 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,445 | 1,450 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,405 | 1,406 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,476 | 1,471 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,434 | 1,435 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,442 | 1,440 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,421 | 1,422 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 48.2 | 58.6 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 21.4 | 29.2 |
| Artificial Analysisreported | GPQA Diamond% of questions | 89.3% | 92.9% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 29.0% | 39.0% |
| Artificial Analysisreported | IFBench% of instructions | 78.8% | 82.9% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 77.3% | 83.0% |
| Artificial Analysisreported | Output speedtokens per second | 89.7 | 80.6 |
| Artificial Analysisreported | SciCode% of problems | 44.8% | 47.1% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 51.3% | 65.2% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 0.0% | 2.0% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 40.9% | 42.4% |
| Artificial Analysisreported | Time to first answer tokenseconds | 1.57 | 25.9 |
| Artificial Analysisreported | Τ-bench banking% of tasks | 13.4% | 15.3% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 95.6% | 88.9% |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.
Quick answers
Is Qwen3.5 397B A17B better than MiniMax M3?
It depends on the task. Among models available today, on our benchmarks and licensed sources, Qwen3.5 397B A17B ranks higher for agents and tool use and sticking to the facts; MiniMax M3 ranks higher for professional work, reasoning and knowledge and writing.
Which is cheaper, Qwen3.5 397B A17B or MiniMax M3?
MiniMax M3 costs $0.52 per million tokens (three input to one output), against $1.29 for Qwen3.5 397B A17B.
Which is faster, Qwen3.5 397B A17B or MiniMax M3?
Qwen3.5 397B A17B writes about 90 tokens a second on its usual API, against 81 for MiniMax M3 (Artificial Analysis).
Run Qwen3.5 397B A17B and MiniMax M3 on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.