Artificial Analysis describes a benchmark developed with Google DeepMind for agents handling visually complex inputs and creating document deliverables across 200 tasks in 10 industries. Scores and cost estimates depend on its rubrics and evaluation harness; benchmark development is a partnership, not wholly independent of the model maker.
Preprint, not yet peer-reviewed. Introduces a visual harness that lets multimodal models retain and retrieve previous observations during interactive tasks. The authors report improvements on visual games and puzzles with the same underlying model; these are research benchmark results, not independent production validation.
MiniMax’s public model card and downloadable weights for a multimodal mixture-of-experts model supporting text, image, and video inputs. The company describes coding, agent workflows, computer use, and cybersecurity applications and a context window up to one million tokens; its benchmark results are company-reported. Noncommercial use is free under the MiniMax Community License. Commercial use requires attribution and notice, with prior authorization above the license’s revenue threshold; running the model requires suitable computing hardware.
Xiaomi's open-weight AI model family — text, image, video, and audio understanding — released under the MIT license, free to download and self-host.
Why I recommend it: Weights are free (MIT license, commercial use allowed) on Hugging Face. The hosted API is paid per token. If you can run models locally, this is a free frontier-tier option; if you want a chat interface, use the API and expect a bill.
PrismML builds 1-bit and ternary Bonsai models that pack advanced reasoning, tool use and multimodal AI into efficient devices and data centers.
From the site: PrismML builds 1-bit and ternary Bonsai models that pack advanced reasoning, tool use, and multimodal AI into phones, laptops, and efficient data centers.
Why I recommend it: Verify current pricing and free-tier availability on site; the homepage focuses on enterprise and developer access.
#efficient AI#1-bit models#Bonsai#multimodal#edge AI