Qwen Team: native vision-language hybrid training resource

What changed

Combined early multimodal fusion with alternating gated-delta and global attention layers in a sparse expert model family, releasing weights and implementation-facing architecture details.

Scientific scope

Combined early multimodal fusion with alternating gated-delta and global attention layers in a sparse expert model family, releasing weights and implementation-facing architecture details. Integrated native vision-language/hybrid training resource, not invention of GatedDeltaNet, sparse experts or asynchronous RL. Architecture and creator citation are documented, but training-efficiency marketing figures are not independently certified. Sizes and later same-architecture releases are not separate inventions.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Integrated native vision-language/hybrid training resource, not invention of GatedDeltaNet, sparse experts or asynchronous RL. Architecture and creator citation are documented, but training-efficiency marketing figures are not independently certified. Sizes and later same-architecture releases are not separate inventions.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Qwen3.5: Towards Native Multimodal Agents — official model card and creator citation

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction