Agnes AI releases Agnes-3.0-Flash, a 33B open‑weights multimodal model with tool calling
Agnes AI announced the open‑weights release of Agnes-3.0-Flash, a 33‑billion‑parameter multimodal model that handles text, image and video inputs while offering built‑in tool‑calling and adjustable reasoning effort. The model uses a hybrid‑attention decoder where three out of four layers run a gated‑delta recurrent rule, leaving only 18 of the 72 layers with a KV cache, which reduces memory…
Key points
- Agnes-3.0-Flash is a 33B multimodal model supporting text, image, video, and tool‑calling.
- Hybrid‑attention decoder reduces KV‑cache layers to 18, enabling a 144‑token context window on cheaper hardware.
- Released under Apache 2.0 with Hugging Face integration and Docker serving scripts.
The team provided a full Hugging Face‑compatible pipeline, including a custom chat template that lets developers select high, medium or low reasoning levels, toggle thinking off, and embed tool definitions that the model emits as structured calls. Sample Docker commands show how to serve the model with sglang, and the release is licensed under Apache 2.0 with a citation pointing to https://agnes‑ai.com/. While the benchmark table is a reference compilation rather than a controlled head‑to‑head test, the authors claim competitive results on reasoning, coding and instruction‑following tasks, positioning Agnes‑3.0‑Flash as a cost‑effective alternative to larger flagship models.
Agnes-AI/Agnes-3.0-Flash 33B Multimodal, AA score: 36
huggingface.co · 12 September 2026
Loading the full article…
This text was published by huggingface.co. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Google launches Gemini 4 Argon model globally for Pro and Ultra subscribers with introductory API pricing · 32 src
- Meta releases Muse Spark Open, an 816B-parameter open-weight multimodal model with 42B active parameters and 1M-token context · 20 src
- Microsoft AI chief says in-house model beats OpenAI's using quarter of compute · 1 src
- NVIDIA releases Nemotron 3.5 Super VL and Bolt edge models for enterprise agents · 8 src
- Black Forest Labs releases FLUX 3 Image Klein, a 5.6B-Parameter Open-Weights model for text-to-image generation · 3 src
Comments
via GitHub Discussions