Alibaba
CoverageBenchmark
Alibaba's Qwen3.6-27B, released April 22, 2026, outperforms the 397B Qwen3.5-397B-A17B on agentic coding benchmarks and runs at 80 tokens per second on a
Get 10-25% off from Qwen
Model details
Qwen3.5 397B-A17B
Released by Alibaba's Qwen team as the first open-weight entry in the Qwen3.5 series, Qwen3.5 397B-A17B is positioned as a native vision-language model that combines broad multimodal understanding with strong reasoning, coding, and agent capability. According to the Qwen team's launch announcement, the model is built on an innovative hybrid architecture that fuses linear attention (via Gated Delta Networks) with a sparse mixture-of-experts design, allowing it to reach 397 billion total parameters while activating only 17 billion per forward pass. That structure is intended to deliver higher inference efficiency without giving up capability, making the model practical for developers and enterprises that need multimodal reasoning at scale.
Qwen3.5 397B-A17B fits naturally into multimodal agent workflows, where a single model must read images and text, reason about them, write code, and call tools. The Qwen team reports outstanding results across benchmark suites covering reasoning, coding, agent tasks, and multimodal understanding, while language coverage was expanded from 119 to 201 languages and dialects for broader global reach. Official weights and code are published on Hugging Face, ModelScope, and GitHub under the Qwen organization, so teams can self-host, fine-tune, or integrate the model into downstream pipelines rather than depending solely on hosted endpoints.
Alibabaqwen3.5-397b-a17bqwen
Quick Info
Powered by- Provider
- Alibaba
- Model key
- qwen3.5-397b-a17b
- Release date
- Feb 15, 2026
- Last updated
- Feb 15, 2026
- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $0.60
- Output token cost
- $3.60
Limits
- Output tokens
- 65,536 tokens
- Context window
- 262,144 tokens
Transparent token rates
Compare Qwen3.5 397B-A17B pricing
Rates are shown per one million tokens. Combined means one million input plus one million output tokens.
Latest news about Qwen3.5 397B-A17B
Alibaba
Coverage
Baidu Baike documents Qwen3.5 as a flagship large language model series launched by Alibaba on February 16, 2026, comprising two versions: Qwen3.5-Plus and Qwen3.5-397B-A17B. The series uses a Hybrid Attention Mechanism and Sparse Mixture of Experts architecture, optimized for logical reasoning, mathematical computatio The encyclopedia entry traces the development timeline, noting Alibaba announced Qwen3.5's development plan in January 2026, with a code merge request appearing on Hugging Face on February 9, followed by showcases at major model events on February 11. International hardware makers including NVIDIA, AMD, and Apple achie
Cloudflare AI Gateway
CoverageBenchmark
SemiAnalysis's InferenceX provides an independent technical breakdown of Qwen3.5-397B-A17B, confirming the model has 397B total parameters with 17B activated per forward pass and was publicly released in mid-February 2026. Weights are published on Hugging Face under Apache License 2.0 as a post-trained checkpoint compa The article reports an Artificial Analysis Intelligence Index score of 45, ranking third among open-weights models at publication, with a GDPval-AA agentic ELO of 1,221 versus 860 for Qwen3 235B. Training uses multi-token prediction (MTP) across multiple steps. The model achieves cross-generational parity with Qwen3 wh