Sage-A1.5 Brings Agentic Vision to Open MoE Models

InclusionAI has published Sage-A1.5, a vision-language model that leans into agentic behavior rather than static image captioning. Released on Hugging Face under the permissive Apache 2.0 license, the model combines multimodal understanding with reasoning, tool-use, and code capabilities — a combination aimed squarely at systems that need to see, plan, and act.

The model uses a mixture-of-experts (MoE) architecture, a design that activates only a subset of parameters per token to keep inference costs lower than a comparably sized dense model. That approach has become increasingly common among teams trying to scale capability without proportionally scaling compute.

Why it matters

Agentic vision-language models are a growing category because many real-world tasks — navigating interfaces, interpreting screenshots, or debugging visual output — require a model that can both perceive and take action. Sage-A1.5 folds several of these capabilities into one release:

• Vision-language understanding as the primary modality
• Reasoning and tool-use for multi-step tasks
• Code generation for executable workflows

The open license is the other notable piece here. Apache 2.0 allows commercial use and modification, lowering the barrier for developers who want to build agents on top of an open base rather than a closed API.

InclusionAI has not published detailed parameter counts or context-length figures alongside this release, so builders will want to consult the model card directly for specifics before integrating it into production pipelines.