Researchers introduced AXPO, a policy optimization method that fixes the Thinking-Acting Gap in vision-language models by improving tool utilization rates from ~30% to higher success rates through thinking prefix optimization and tool call resampling. For commerce practitioners building AI agents, this means more reliable autonomous tool use in product search, inventory queries, and customer service workflows without scaling model size.

Multimodal AI models scale toward unified vision-language systems

NEO-ov native vision-language model unifies pixel-to-word learning at scale

LLM

Researchers published NEO-ov, a native vision-language model that learns cross-frame and pixel-word correspondences end-to-end without modular components, achieving competitive performance on visual perception tasks. For commerce practitioners, this unified architecture enables more efficient multimodal AI for product understanding, video analysis, and spatial reasoning without the latency penalties of stitched-together encoder-decoder systems.

May 28, 2026

Multimodal AI models scale toward unified vision-language systems

NVIDIA Gamma-World scales multi-agent video generation to four players.

Entertainment / Recreation

NVIDIA researchers introduced Gamma-World, a generative multi-agent world model using Simplex Rotary Agent Encoding and Sparse Hub Attention to enable real-time interactive video generation with multiple controllable agents at 24 FPS, generalizing from two to four players without retraining. Commerce platforms building multiplayer simulations, virtual showrooms, or interactive product demonstrations can now generate consistent, action-responsive environments with multiple participants at scale, reducing computational overhead from quadratic to linear attention complexity.

May 28, 2026

AI in Commerce News

AXPO improves vision-language agent tool use and reasoning

NEO-ov native vision-language model unifies pixel-to-word learning at scale

NVIDIA Gamma-World scales multi-agent video generation to four players.