# Agnes-3.0-Flash arrives as a multimodal reasoning model

> The new release pairs vision-language understanding with a hybrid-attention design aimed at long-context reasoning.

Published by The Open Weights on Sep 11, 2026. Canonical: https://theopenweights.com/news/agnes-3-0-flash-z86d

## Key facts

- Company: Agnes AI
- Model: Agnes-3.0-Flash
- Version: 3.0-Flash
- Category: Vision-Language
- Modalities: Vision-Language, Reasoning, Text / LLM
- License: Other (Open weights, commercial use allowed)
- Context window: 262K tokens
- Architecture: AgnesForConditionalGeneration
- Precision: BF16
- Weights size: 66.2 GB
- Significance: minor
- Published: 2026-09-11
- Last verified: 2026-09-13
- Hugging Face: https://huggingface.co/Agnes-AI/Agnes-3.0-Flash
- Canonical URL: https://theopenweights.com/news/agnes-3-0-flash-z86d

Agnes-3.0-Flash has landed on Hugging Face, positioned as a multimodal model that combines vision-language understanding with a dedicated reasoning mode. According to its [model page](https://huggingface.co/Agnes-AI/Agnes-3.0-Flash), the release leans on a hybrid attention scheme and support for long context, two ingredients increasingly common in the current generation of efficiency-minded models.

The "Flash" label signals the familiar trade-off in today's model lineups: a variant tuned for speed and cost rather than maximum capability. Models in this tier are typically meant to handle everyday multimodal tasks — reading images, parsing documents, and answering questions — while keeping latency and compute demands modest.

## What we know

- Primary modality is vision-language, with additional text and reasoning capabilities.
- The architecture uses hybrid attention and targets long-context workloads.
- It is distributed under a custom license listed simply as "other."

Several key details remain unstated in the public record, including parameter count, exact context window, and benchmark results. That makes it hard to place Agnes-3.0-Flash precisely against comparable open releases, and prospective users will want to check the license terms closely before building on it.

**Why it matters:** Compact multimodal models with long context are becoming the workhorses of practical AI deployments, and each new entrant widens the field of options for developers who need vision and reasoning without frontier-scale costs. The full specifications are available on the [Hugging Face repository](https://huggingface.co/Agnes-AI/Agnes-3.0-Flash).

## Get the model

- [Hugging Face](https://huggingface.co/Agnes-AI/Agnes-3.0-Flash)

## Sources

- [Agnes-AI/Agnes-3.0-Flash](https://huggingface.co/Agnes-AI/Agnes-3.0-Flash) — Hugging Face, Sep 11, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Agnes-3.0-Flash arrives as a multimodal reasoning model", The Open Weights, Sep 11, 2026, https://theopenweights.com/news/agnes-3-0-flash-z86d