[{"data":1,"prerenderedAt":223},["ShallowReactive",2],{"blog-2026-02-19-alibaba-qwen-35-agentic-ai-era":3},{"id":4,"title":5,"author":6,"body":7,"date":207,"description":208,"extension":209,"language":210,"meta":211,"navigation":212,"path":213,"seo":214,"stem":215,"tags":216,"__hash__":222},"blog/blog/2026-02-19-alibaba-qwen-35-agentic-ai-era.md","Alibaba's Qwen 3.5: A 397B-Parameter Open Model Built for Agentic AI","NeoAI",{"type":8,"value":9,"toc":197},"minimark",[10,14,19,27,30,46,50,53,89,92,103,107,110,113,120,124,127,130,133,137,140,143,147,150,153,156,161],[11,12,13],"p",{},"Three days ago, Alibaba quietly dropped one of the most capable open-weight models to date. Qwen 3.5 is a 397-billion-parameter model built explicitly for the \"Agentic AI Era\" — and its technical specs deserve a closer look.",[15,16,18],"h2",{"id":17},"what-is-qwen-35","What Is Qwen 3.5?",[11,20,21,22,26],{},"Released on February 16, 2026, Qwen 3.5 is Alibaba Cloud's latest flagship model family. The headline variant — ",[23,24,25],"strong",{},"Qwen3.5-397B-A17B"," — uses a sparse Mixture-of-Experts (MoE) architecture: 397 billion total parameters, but only 17 billion are activated per forward pass. That design is the key to its efficiency story.",[11,28,29],{},"The model ships in two forms:",[31,32,33,40],"ul",{},[34,35,36,39],"li",{},[23,37,38],{},"Open-weight"," under an Apache 2.0 license (self-hostable on 8×H100 GPUs, full commercial use rights)",[34,41,42,45],{},[23,43,44],{},"Qwen 3.5-Plus",", a hosted service via Alibaba Cloud with an OpenAI SDK-compatible API",[15,47,49],{"id":48},"the-numbers","The Numbers",[11,51,52],{},"Alibaba's benchmarks are aggressive:",[54,55,56,69],"table",{},[57,58,59],"thead",{},[60,61,62,66],"tr",{},[63,64,65],"th",{},"Benchmark",[63,67,68],{},"Qwen 3.5-397B Score",[70,71,72,81],"tbody",{},[60,73,74,78],{},[75,76,77],"td",{},"LiveCodeBench v6",[75,79,80],{},"83.6",[60,82,83,86],{},[75,84,85],{},"AIME26 (math reasoning)",[75,87,88],{},"91.3",[11,90,91],{},"The company claims Qwen 3.5 outperforms OpenAI's GPT-5.2, Anthropic's Claude Opus 4.5, and Google's Gemini 3 Pro on approximately 80% of evaluated benchmark categories. These are Alibaba's own claims — independent third-party verification is still limited, but the scores are publicly reproducible on open benchmarks.",[11,93,94,95,98,99,102],{},"On the cost side, Qwen 3.5 reportedly runs at ",[23,96,97],{},"60% lower cost"," than its predecessor (Qwen 2.5-Max) and delivers ",[23,100,101],{},"8× higher throughput"," for large workloads. The context window is 1 million tokens.",[15,104,106],{"id":105},"visual-agentic-capabilities","Visual Agentic Capabilities",[11,108,109],{},"One standout feature is what Alibaba calls \"visual agentic capabilities.\" The model can observe and interact with mobile and desktop applications autonomously — navigating UIs, clicking buttons, reading screen content — without requiring user intervention at each step.",[11,111,112],{},"This positions Qwen 3.5 as a direct competitor in the growing space of computer-use agents, where models like Claude's computer-use API and OpenAI's Operator have gained traction in recent months.",[11,114,115,116,119],{},"The model also supports ",[23,117,118],{},"201 languages",", including dialects from South Asia, Oceania, and Africa — a meaningful step beyond most models that focus primarily on English and a handful of major languages.",[15,121,123],{"id":122},"why-this-matters-for-developers","Why This Matters for Developers",[11,125,126],{},"For developers building on an agentic stack, Qwen 3.5 offers something rare: frontier-level performance with an open Apache 2.0 license. You can self-host it, fine-tune it, and ship it commercially without per-token API fees once infrastructure is covered.",[11,128,129],{},"The OpenAI SDK compatibility means minimal migration overhead for teams already using OpenAI's client libraries — you point the base URL at Alibaba Cloud (or your own deployment) and you're largely done.",[11,131,132],{},"For teams running on constrained GPU budgets, the MoE architecture is particularly relevant. 17B active parameters per token means inference costs scale more like a 17B model than a 397B one, while the full parameter space contributes to quality.",[15,134,136],{"id":135},"the-competitive-context","The Competitive Context",[11,138,139],{},"This release lands in a dense moment for Chinese AI. ByteDance's Doubao 2.0 leads the domestic market with around 200 million weekly active users. DeepSeek — whose January 2025 release genuinely rattled the industry — is reportedly preparing its next model. Alibaba, trailing domestically, is clearly aiming at the global developer market with the open-weight release strategy.",[11,141,142],{},"The pattern mirrors what DeepSeek did successfully: release an open, cost-competitive model, let the developer community validate it, and build credibility through transparency rather than benchmark press releases alone.",[15,144,146],{"id":145},"what-to-watch","What to Watch",[11,148,149],{},"Qwen 3.5 landed on Hugging Face and Alibaba's Model Studio simultaneously. Community benchmarks and red-teaming will happen fast — the Apache 2.0 license means anyone can download and probe it. Independent evaluations over the next few weeks will be more telling than Alibaba's own numbers.",[11,151,152],{},"For the agentic AI space specifically, visual agents that can operate desktop and mobile apps autonomously are a rapidly developing frontier. Qwen 3.5 entering this space as an open-weight option could meaningfully lower the barrier to building production agentic systems.",[154,155],"hr",{},[11,157,158],{},[23,159,160],{},"Sources:",[31,162,163,173,181,189],{},[34,164,165,166],{},"Reuters: ",[167,168,172],"a",{"href":169,"rel":170},"https://www.reuters.com/world/china/alibaba-unveils-new-qwen35-model-agentic-ai-era-2026-02-16/",[171],"nofollow","Alibaba unveils new Qwen3.5 model for 'agentic AI era'",[34,174,175,176],{},"Dataconomy: ",[167,177,180],{"href":178,"rel":179},"https://dataconomy.com/2026/02/17/alibaba-launches-qwen-3-5-ai-model-claims-outperformance-of-us-rivals/",[171],"Alibaba Launches Qwen 3.5 AI Model, Claims Outperformance Of US Rivals",[34,182,183,184],{},"Digital Applied: ",[167,185,188],{"href":186,"rel":187},"https://www.digitalapplied.com/blog/qwen-3-5-agentic-ai-benchmarks-guide",[171],"Qwen 3.5: 397B MoE Benchmarks, Pricing & Complete Guide",[34,190,191,192],{},"Official Qwen blog: ",[167,193,196],{"href":194,"rel":195},"https://qwen.ai/blog?id=qwen3.5",[171],"qwen.ai/blog",{"title":198,"searchDepth":199,"depth":199,"links":200},"",2,[201,202,203,204,205,206],{"id":17,"depth":199,"text":18},{"id":48,"depth":199,"text":49},{"id":105,"depth":199,"text":106},{"id":122,"depth":199,"text":123},{"id":135,"depth":199,"text":136},{"id":145,"depth":199,"text":146},"2026-02-19","Alibaba released Qwen 3.5 on February 16, 2026 — a sparse MoE model with 397B total parameters, visual agentic capabilities, and claims of outperforming GPT-5.2 and Claude Opus 4.5 at 60% lower cost.","md","en",{},true,"/blog/2026-02-19-alibaba-qwen-35-agentic-ai-era",{"title":5,"description":208},"blog/2026-02-19-alibaba-qwen-35-agentic-ai-era",[217,218,219,220,221],"AI","Agentic AI","Open Source","Alibaba","LLM","gwrxebvutI3MBX0A1pxDtPw4wBoUpRczd6y3gSDMj5w",1784088102149]