Qwen3.5 397B A17B
AlibabaReleased Feb 15, 2026Open weights · Apache 2.0qwen3.5-397b-a17b
- Intelligence (Epoch)
- 147 · 50th of 111
- Reasoning
- yes
- Knowledge cutoff
- September 2025
- Context
- 262K tokens
- Max output
- 82k tokens
- Input
- text, image, video
- API features
- tools, structured outputs, JSON mode, seed, logprobs
- Alibaba's API
- yes, as qwen3.5-397b-a17b
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from Alibaba · cutoff: aiknowledgecutoff.com
The cheapest host, DeepInfra, serves fp8 weights (maker: bf16). Weights and context per host
Intelligence for the price
14 models score higher for the same price or less; it ranks 15 of the 69 models at its price or below.
Epoch Capabilities Index against the lowest input + output price per M, log scale · prices Oct 23. Score: Epoch AI, CC BY 4.0, 147 with an interval of 145 to 149.
Against its rivals
The best of its price range and the top score
- Qwen3.5 397B A17Bthis model$3.45147
- Claude Opus 5.5top score$24.00169
- Gemini 3.7 Flashsame price range$4.50158
- Gemini 3.8 Flashsame price range$4.50157
Same level for 22× less: Qwen3.7 Flash at Alibaba, $0.16 per M input + output · their Epoch intervals overlap (144 against 147).
Bars: Epoch Capabilities Index (Epoch AI, CC BY 4.0), drawn from 100, below the lowest score in the catalogue · price: lowest input + output per M.
Speed by provider
Output speed of Qwen3.5 397B A17B at each host, fastest first
- StreamLake52 tok/s1.56 s to first token
- GMICloud51 tok/s4.90 s to first token
- Venice49 tok/s4.09 s to first token
- DeepInfra48 tok/s0.93 s to first token
- AtlasCloud48 tok/s2.56 s to first token
- Alibaba47 tok/s0.89 s to first token
- Phala44 tok/s0.87 s to first token
- DigitalOcean21 tok/s1.63 s to first token
- Parasail21 tok/s2.07 s to first token
Speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · a snapshot of real traffic before the daily read, not a daily average · bars from zero.
Run it yourself
open weights, Apache 2.0 · fits 1 of the 6 workstations in our guideSelf-host guide ›- Parameters
- 397B
- 17B active per token (MoE)
- Memory needed
- 238 GB
- in INT4 · 476 GB in INT8
- Smallest cloud setup
- 4× H100
- 320 GB · $8.60/h, about $6,278 a month
- Worth it above
- 2B tokens
- a month, against the cheapest API at $3.45 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 238 GB | 4× H100 · $8.60/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 476 GB | 8× H100 · $17.20/h | near-lossless |
| FP16 (full) | 953 GB | multi-node, over 640 GB | reference quality |
At home: in INT4 it fits a Mac Studio M5 Ultra (256 GB unified), about $9,499 once, or $0.3615/h over 3 years, electricity aside; it pays for itself after about 3B tokens against the cheapest API.
Memory ≈ parameters × bytes per parameter × 1.2 · cloud price: on-demand rate, running 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| DigitalOcean128K ctx (256K) | $0.55 | $3.50 | $0.11 | 21 | 1.63 s | 99.2% | Oct 23 |
| StreamLakeprecision not read: host terms · 250K ctx · 64K out | $0.60 | $3.60 | $0.12 | 52 | 1.56 s | 98.3% | Oct 23 |
| DeepInfrafp8 (bf16) · 256K ctx · 80K out | $0.45 | $3.00 | $0.22 | 48 | 0.93 s | 100% | Oct 23 |
| Phalaprecision not read: host terms · 256K ctx | $0.55 | $3.50 | $0.23 | 44 | 0.87 s | 98.9% | Oct 23 |
| Parasailfp8 (bf16) · 256K ctx | $0.50 | $3.60 | $0.30 | 21 | 2.07 s | 100% | Oct 23 |
| AtlasCloudfp8 (bf16) · 256K ctx · 64K out | $0.55 | $3.50 | $0.55 | 48 | 2.56 s | 98.8% | Oct 23 |
| Alibabaprecision not read: host terms · 256K ctx · 64K out | $0.60 | $3.60 | — | 47 | 0.89 s | 100% | Oct 23 |
| GMICloudfp8 (bf16) · 256K ctx | $0.60 | $3.60 | — | 51 | 4.90 s | 100% | Oct 23 |
| Novita AIprecision not stated · 256K ctx · 64K out | $0.60 | $3.60 | — | — | — | — | Oct 23 |
| Veniceprecision not read: host terms · 125K ctx (256K) · 32K out | $0.75 | $4.50 | — | 49 | 4.09 s | 82.3% | Oct 23 |
10 of 10 providers · prices per million tokens, read Oct 23, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| DeepInfra | fp8 | 256K | 80K | yes | yes | yes |
| DigitalOcean | not checked yet | 128K | not checked yet | yes | yes | yes |
| Phala | not read: host terms | 256K | not read: host terms | yes | yes | yes |
| AtlasCloud | fp8 | 256K | 64K | yes | yes | yes |
| Parasail | fp8 | 256K | not read: host terms | yes | yes | yes |
| Alibaba | not read: host terms | 256K | 64K | yes | no | yes |
| GMICloud | fp8 | 256K | not read: host terms | yes | no | yes |
| Novita AI | not stated by the host | 256K | 64K | yes | no | yes |
| StreamLake | not read: host terms | 250K | 64K | no | yes | yes |
| Venice | not read: host terms | 125K | 32K | yes | yes | yes |
Weights, context and max output: read at each host's own catalogue or docs where its terms allow, otherwise as the host declares to OpenRouter; hover a cell for its source and date · tools, structured outputs, reasoning: as declared to OpenRouter · not verified by us
Qwen3.5 397B A17B spec sheet, field by field, with sources
Qwen3.5 397B A17B spec sheetmaker-declared
What Alibaba declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.
- Context window
- 262,144 tokens
- Parameters
- 403,397,928,944≈ 403.4B
- Active experts
- 10 of 512 active per token
- Weight precision
- BF16
- License
- Apache 2.0 — open weights, permissivecurated · read 2026-09-07Scope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ September 2025 third-party estimate — not the maker's figurecurated
- Modalities
- text · image · video
- Max output (official)
- to verify
- Release date
- to verify≈ February 15, 2026 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- Qwen3_5MoeForConditionalGeneration
- Layers
- 60
- Vocabulary
- 248,320 tokens
- Weights first published
- February 16, 2026
- Weights last updated
- April 24, 2026
Qwen/Qwen3.5-397B-A17B · config.json @ 8472618 · read 2026-10-08 · 6 of 10 fields published by the maker
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| DeepInfra | $1.95 | $5.10 | $4.80 |
| DigitalOcean | $2.30 | $6.15 | $5.70 |
| Phala | $2.30 | $6.15 | $5.70 |
| AtlasCloud | $2.30 | $6.15 | $5.70 |
| Parasail | $2.30 | $5.80 | $5.60 |
| Alibaba | $2.40 | $6.60 | $6.00 |
| GMICloud | $2.40 | $6.60 | $6.00 |
| Novita AI | $2.40 | $6.60 | $6.00 |
| StreamLake | $2.40 | $6.60 | $6.00 |
| Venice | $3.00 | $8.25 | $7.50 |
What each host bills for 1,000 requests of a representative job
Your own workload
Other Alibaba models · all its models, hosts and history ›
- Qwen3.8 27B16 providers · from $1.29
- Qwen3.6 35B A3B12 providers · from $1.00
- Qwen3 235B A22B Instruct 250710 providers · from $0.44
- Qwen3.6 27B9 providers · from $2.25
- Qwen3.5-35B-A3B8 providers · from $1.14
- Qwen3 30B A3B7 providers · from $0.24