Qwen3.5 397B-A17B: benchmark scores, price and ranking
Qwen3.5 397B-A17B from Alibaba ranks #146 of 743 LLMs on the LLMs Tiger aggregate benchmark score (51.1), based on 10 benchmark results captured up to 2026-10-13.
- Organisation
- Alibaba
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2026-02-13 (source: epoch)
- Cheapest API price
- $0.164 in / $0.986 out per million tokens (via aihubmix, observed 2026-10-13T19:47Z)
- Aggregate rank
- #146 of 743 (score 51.1)
Relative to other models, Qwen3.5 397B-A17B is strongest on MMLU-Pro (#4 of 136) and weakest on APEX-Agents (#14 of 16).
Across the 10 benchmarks where at least ten models are scored, Qwen3.5 397B-A17B is in the top 10% on 2 and in the top half on 7.
Its cheapest listed input price, $0.164 per million tokens, is lower than that of 71% of the 427 models with a listed price.
Qwen3.5 397B-A17B benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| MMLU-Pro | 87.8% | #4 of 136 | HuggingFace MMLU-Pro leaderboard | 2026-10-13 |
| Epoch ECI | 147 | #20 of 204 | Epoch AI Benchmarking Hub | 2026-10-13 |
| GPQA Diamond | 88.4% | #45 of 275 | HuggingFace GPQA Diamond leaderboard | 2026-10-13 |
| DTBench | 87.5% | #34 of 146 | Epoch AI Benchmarking Hub | 2026-10-13 |
| OTIS Mock AIME | 88.9% | #48 of 189 | Epoch AI Benchmarking Hub | 2026-10-13 |
| LMCA | 37.9% | #34 of 108 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Chess Puzzles | 13.0% | #60 of 122 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Mystery Games | 18.0% | #39 of 59 | Epoch AI Benchmarking Hub | 2026-10-13 |
| FrontierMath T1-3 | 31.2% | #51 of 68 | Epoch AI Benchmarking Hub | 2026-10-13 |
| APEX-Agents | 24.9% | #14 of 16 | Epoch AI Benchmarking Hub | 2026-10-13 |
About Qwen3.5 397B-A17B
Qwen3.5 397B-A17B is an open-weight model that can run on your own hardware from Alibaba. It was released on 2026-02-13, according to Epoch AI. The cheapest API listing we track charges $0.164 per million input tokens and $0.986 per million output tokens, through aihubmix. We hold 10 benchmark results for it across 3 areas, and 2 are in the top tenth of their boards.
Where Qwen3.5 397B-A17B stands
On our aggregate score, Qwen3.5 397B-A17B is #146 of 743 ranked models, in the top quarter. Its score is 51.1; the scale is centred on 50, so it sits 1.1 points above the typical model. That beats 80% of the ranked field. Among the 74 ranked Alibaba models it is #7. The next one up is Qwen2.5-Coder-14B-Instruct at 51.3. The next one down is Qwen3.8 Max (0902) (xhigh) at 50.8.
Qwen3.5 397B-A17B, benchmark by benchmark
On MMLU-Pro, Qwen3.5 397B-A17B scores 87.8% and ranks #4 of 136. Only 3 models score higher, led by Darwin-180B-RSI at 88.1%, which is 0.3 points ahead. It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Qwen3.5 397B-A17B is 32.5 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.
On Epoch ECI, Qwen3.5 397B-A17B scores 147 and ranks #20 of 204. Only 19 models score higher, led by Qwen 3.8 Max at 156, which is 10 points ahead. It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so Qwen3.5 397B-A17B is 19 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Qwen3.5 397B-A17B manages 88.4% on GPQA Diamond, good for #45 of 275. GPT-6 Astra (max) leads at 95.8%, 7.4 points further up, and 84% of models sit below. The results either side are Darwin-36B-Opus (88.4%) and Claude Opus 4.6 (max) (88.4%). It covers expert-level science questions. The board median is 71.7%, so Qwen3.5 397B-A17B is 16.7 points above it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.
DTBench puts Qwen3.5 397B-A17B at 87.5%, #34 of 146. That beats 77% of the models tested, but 11.5 points short of Claude Opus 5.5 (max) (98.9%). The results either side are Inkling (xhigh) (87.5%) and Qwen 3.6 Max (Preview) (87.2%). It covers dynamic tool-use benchmark. The board median is 68.8%, so Qwen3.5 397B-A17B is 18.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
On OTIS Mock AIME, Qwen3.5 397B-A17B reaches 88.9%. It ranks #48 of 189, ahead of 75% of the field, and 11.1 points behind Claude Fable 5.1 (max) at 100.0%. The results either side are Inkling (xhigh) (88.9%) and GPT-5.1 (high) (88.6%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so Qwen3.5 397B-A17B is 31.1 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Qwen3.5 397B-A17B manages 37.9% on LMCA, good for #34 of 108. Claude Opus 5.5 (max) leads at 68.2%, 30.3 points further up, and 69% of models sit below. The results either side are grok-4.3 (high) (38.3%) and Inkling (xhigh) (37.6%). It covers language-model capability assessment. The board median is 29.2%, so Qwen3.5 397B-A17B is 8.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Chess Puzzles puts Qwen3.5 397B-A17B at 13.0%, #60 of 122. That beats 51% of the models tested, but 59.0 points short of GPT-6 Astra (max) (72.0%). The results either side are GPT-4o (Aug 2024) (13.0%) and seed-oss-36b-instruct (13.0%). It covers chess tactics puzzles. The board median is 12.5%, so Qwen3.5 397B-A17B is 0.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Qwen3.5 397B-A17B lands below the middle on Mystery Games: 18.0%, #39 of 59, with 64% of the field scoring higher. GPT-6 Astra (max) leads at 84.0%; the closest result above is Qwen 3.6 Flash (2026-04-16) at 18.0%. It covers deductive reasoning puzzles. The board median is 23.0%, so Qwen3.5 397B-A17B is 5.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Qwen3.5 397B-A17B lands below the middle on FrontierMath T1-3: 31.2%, #51 of 68, with 74% of the field scoring higher. GPT-6 Astra (max) leads at 93.7%; the closest result above is o3 (high) at 33.3%. It covers research-level maths, tiers 1-3. The board median is 55.6%, so Qwen3.5 397B-A17B is 24.4 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
APEX-Agents is a weak spot. Qwen3.5 397B-A17B scores 24.9%, #14 of 16, behind 81% of the models measured. The leader, Claude Sonnet 5.5 (max), reaches 75.5%. It covers long professional agent tasks. The board median is 47.8%, so Qwen3.5 397B-A17B is 22.9 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Strengths and weaknesses
Reasoning, maths and science: solid, ahead of 70% of the field on average across 4 benchmarks. Its best board here is MMLU-Pro (#4 of 136) and its weakest is FrontierMath T1-3 (#51 of 68). Agents and tool use: uneven, ahead of 45% of the field on average across 2 benchmarks. Its best board here is DTBench (#34 of 146) and its weakest is APEX-Agents (#14 of 16). General knowledge and chat: solid, ahead of 61% of the field on average across 4 benchmarks. Its best board here is Epoch ECI (#20 of 204) and its weakest is Mystery Games (#39 of 59).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so Qwen3.5 397B-A17B costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen3.7 Flash ($0.030, 50.5), GLM-5.2 (high) ($0.060, 51.6), glm-4.7-flash ($0.060, 50.1). Similar scores cost more on grok-4-1-fast-non-reasoning ($0.200) and DeepSeek V3 0324 ($0.200).
Release history
The Alibaba model released before it was qwen3-coder-next on 2026-02-02, which scores 45.5. The next one was Qwen3.5 Plus on 2026-02-16, at 49.6.
Reading these numbers
The 10 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (8), HuggingFace GPQA Diamond leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements.
Models ranked nearby
- #143 Claude 4.5 Sonnet (51.2)
- #144 GPT-5 (high) (51.1)
- #145 Claude Code + Fable 5 (51.1)
- #147 Gemini 2.5 Pro Preview (Jun 2025) (51.0)
- #148 grok-4-1-fast-non-reasoning (51.0)
- #149 Quasar Alpha (51.0)
More from Alibaba
- Qwen 3.6 Max (Preview) (#25, 54.1)
- Qwen3.8 Max (xhigh) (#37, 53.2)
- Qwen3.7 Max (#71, 52.3)
- ollama/qwen2.5-coder:32b (#93, 51.9)
- Qwen3 235B A22B diff, no think, Alibaba API (#122, 51.5)
- Qwen2.5-Coder-14B-Instruct (#135, 51.3)
Compare it on the live leaderboard → · How the aggregate score works