Qwen3.5 397B-A17B vs Qwen3.8-Max
| Spec | Qwen3.5 397B-A17B | Qwen3.8-Max |
|---|---|---|
| Respan Index | 29.5 (#93) | 52.9 (#34) |
| Blended price per 1M | $0.387 | $2.48 |
| Context window | 262K | 1M |
| Input / output per 1M | $0.172 / $1.03 | $1.65 / $4.95 |
| Max output | - | 131K |
| Input | text, vision | text, vision |
| Released | 2026-02-15 | 2026-08-02 |
Respan Index by category
Each category is scored 0 to 100, and how far ahead the winner is. Leads under a point are close to even.
| Category | Qwen3.5 397B-A17B | Qwen3.8-Max |
|---|---|---|
| Coding | 11.2 Qwen3.5 397B-A17B | 39.3 Qwen3.8-Max |
| Agents | 34.3 Qwen3.5 397B-A17B | 68.7 Qwen3.8-Max |
| Reasoning | 30.8 Qwen3.5 397B-A17B | 49.4 Qwen3.8-Max |
| Knowledge & language | 27.6 Qwen3.5 397B-A17B | 48.1 Qwen3.8-Max |
| Human preference | 51.1 Qwen3.5 397B-A17B | 58.5 Qwen3.8-Max |
Benchmarks both report
Qwen3.8-Max leads on 10 of the 11 benchmarks both report. Higher is better. Each score links to where it was published.
| Benchmark | Qwen3.5 397B-A17B | Qwen3.8-Max |
|---|---|---|
| LMArena Elo | 1441 | 1484 |
| LMArena WebDev (LMArena) | 1400 | 1671 |
| LMArena Vision (LMArena) | 1246 | 1299 |
| SAGE 1.1 (Vals AI) | 85.5% | 81.8% |
| APEX-Agents (Mercor) | 24.9% | 63.3% |
| Chess Puzzles (Epoch AI) | 13% | 46% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 31.2% | 74.7% |
| GPQA Diamond | 88.4% | 92.6% |
| Mystery Game Puzzles (Epoch AI) | 18% | 38% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 88.9% | 100% |
| tau2-bench Banking Knowledge (Sierra) | 9.8% | 55.2% |
FAQ
Is Qwen3.5 397B-A17B better than Qwen3.8-Max?
Qwen3.8-Max scores 52.9 on the Respan Index (#34), ahead of Qwen3.5 397B-A17B at 29.5 (#93). Qwen3.8-Max leads on 10 of the 11 benchmarks both report.
Which is cheaper, Qwen3.5 397B-A17B or Qwen3.8-Max?
Qwen3.5 397B-A17B lists at $0.172 input and $1.03 output per 1M tokens; Qwen3.8-Max at $1.65 and $4.95. Qwen3.5 397B-A17B is 6.4x cheaper per token at list price.
Which has the larger context window?
Qwen3.5 397B-A17B accepts 262K tokens and Qwen3.8-Max 1M tokens.