Qwen3.5-397B-A17B launches native vision-language model

// 45d agoMODEL RELEASE

Qwen3.5-397B-A17B launches native vision-language model

Qwen3.5-397B-A17B is the first model in Qwen’s Qwen3.5 series and is described by Qwen as a native vision-language model. Qwen has not publicly disclosed a Qwen3.5 training-token total, so the 42–48 trillion estimate should be treated as an inference, not an official figure.

// ANALYSIS

The official evidence supports that Qwen3.5 is a native multimodal successor to Qwen3, but not a public training-token estimate. Qwen’s Qwen3 release documents about 36 trillion tokens; the Qwen3.5 release does not publish a token count, so the 42–48 trillion figure is speculative.

// TAGS

qwenalibabaqwen3qwen3.5multimodalvision-languagellmopen-source

DISCOVERED

45d ago

2026-04-20

PUBLISHED

45d ago

2026-04-20

RELEVANCE

7/ 10

AUTHOR

9r4n4y

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH1h ago

Perplexity is rolling out its agentic "Personal Computer" AI digital worker to Windows machines.

Perplexity has announced the rollout of "Personal Computer for Windows," an agentic AI system designed to operate natively on Windows desktops. Unlike standard chatbot interfaces, this always-on digital worker acts as an autonomous project manager, executing multi-step workflows in a cloud-based environment. It integrates directly with local files and Microsoft 365 applications—including Word, Excel, PowerPoint, Outlook, and OneDrive—to automate tasks. The tool is initially rolling out to paid Max and Enterprise Max subscribers on the waitlist, utilizing a swarm of specialized AI models to complete work in parallel.

NEWS2h ago

Meta has repeatedly delayed the developer API release for its new Muse Spark AI model due to software bugs and infrastructure challenges.

Meta has repeatedly postponed the release of the application programming interface (API) for Muse Spark, its new proprietary multimodal foundation model developed by Meta Superintelligence Labs. Although the model was unveiled in April 2026, integration at scale for developers has been delayed by nearly two months because of technical bugs and infrastructure issues. A Meta spokesperson confirmed that testing is currently underway with a limited set of partners, and the company still intends to release the API to a wider developer audience sometime in June 2026.

MODEL2h ago

Reve 2.0 debuts layout-based image generation

Reve 2.0, developed by @reve, introduces a layout-based image generation system that departs from traditional prompt-to-pixel models. The model uses a learned layout representation combined with pixel diffusion to achieve high-resolution output with precise spatial control.