Qwen3.5-397B-A17B

Name: Qwen3.5-397B-A17B
Author: Alibaba

Released

2026-02-16

Last refreshed

2026-10-09

Status

Researched 134d ago

Open sourceCommercial use: permittedMultimodalCodingRAGAgentsLong contextVisionClassificationJSON / Tool use

Alibaba releases · 51 in the last 12 months · this family litChangelog →

Specifications

Family: Qwen3.5
Released: 2026-02-16
Context: 256K
Parameters: 397B
Architecture: Mixture of Experts
Openness: Open source
License: Apache 2.0OSI-approvedCommercial use: permitted
Weights: Available
Code: Unknown

Created by

Alibaba

AI research institute of Alibaba Group.

Hangzhou, Zhejiang, China

Founded 2017

Website

Pricing

Output / 1M

$2.34

Input / 1M

$0.390

Cheapest of 4 routes · Alibaba Cloud PAI-EAS

Providers(4)

OpenRouter Together AI Alibaba Cloud PAI-EAS Novita AI

View 4 provider routes

About

Alibaba's largest Qwen3.5 model, featuring a Mixture-of-Experts architecture with 397B total parameters and 17B active per token (using 512 total experts with 10 routed + 1 shared active). Supports 201 languages with a native 262K token context window extensible to 1M tokens via YaRN. Includes a thinking/reasoning mode, tool calling with MCP integration, and unified vision-language capabilities through early fusion training.

Providers(4)

Compare all 4

Compare API pricing across 4 tracked provider routes for input and output tokens, batch, and cached reads when available.

Provider	Input / 1M	Output / 1M	Route
Alibaba Cloud PAI-EAS	$0.390	$2.34	Serverless
OpenRouter	$0.390	$2.34	Serverless
Novita AI	$0.600	$3.60	Serverless
Together AI	$0.600	$3.60	Serverless

Available via routers & gateways(1)

OpenRouter

Hybrid

Unified hybrid gateway to 400+ models from 60+ providers via a single OpenAI-compatible API, with optional auto-routing that selects the best model per prompt.

PassthroughTogether AI

Capabilities

VisionMultimodalReasoningJSON / Tool useStructured Outputs

Benchmark peer barsfor Coding

SWE-bench VerifiedRank 40 of 99

Claude Fable 5

96.0

Claude Opus 5

96.0

Claude Mythos Preview

93.9

Claude Opus 4.8

88.6

Qwen3.5-397B-A17Bcurrent

76.2

LiveCodeBenchRank 24 of 73

DeepSeek V4 Pro

93.5

Gemini 3.1 Pro Preview

91.7

DeepSeek V4 Flash

91.6

Qwen3.7-Max

91.6

Qwen3.5-397B-A17Bcurrent

83.6

Benchmark scores(13)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.

Benchmark	Score	Version	Evaluation	Source
Google-Proof Q&A	89.3	diamondObserved 2026-04-18	—	Artificial Analysis
MMLU PRO	87.8	From official HuggingFace model card (accuracy)Observed 2026-06-07	—	Source
Massive Multi-discipline Multimodal Understanding	85.0	—Observed 2026-04-19	—	Source
Instruction-Following Evaluation	92.6	—Observed 2026-04-19	—	Source
BFCL	72.9	v4Observed 2026-04-19	—	Source
SWE-bench Verified	76.2	SWE-bench VerifiedObserved 2026-05-12	—	Source
AIME 2026	91.3	AIME 2026 (accuracy)Observed 2026-06-07	—	Source
Berkeley Function Calling Leaderboard v3	72.9	BFCL-V4, from official model card (accuracy)Observed 2026-06-07	—	Source
Humanity's Last Exam	28.7	HLE with CoT, no tools, from official model card (accuracy)Observed 2026-06-07	—	Source
LiveCodeBench	83.6	LiveCodeBench v6 (pass@1)Observed 2026-06-07	—	Source
MultiChallenge	67.6	Multi-Challenge leaderboard rank 2 of 28 (accuracy%)Observed 2026-06-07	—	Source
τ-bench	86.7	TAU2-Bench, from official model card (accuracy)Observed 2026-06-07	—	Source
Terminal-Bench 2.0	52.5	Terminal-Bench 2.0 (accuracy%)Observed 2026-06-07	—	Source