Hugging Face trending modelspublished Sep 11, 15:50 UTC (4d ago)ingested Sep 13, 14:17 UTCAgnes-AI

Agnes-AI/Agnes-3.0-Flash — new model trending #30 on Hugging Face

AI summary · glm-5.3

Agnes AI releases open-weight Agnes-3.0-Flash Preview, a 33B multimodal model with 262k-token context under Apache 2.0.

Agnes AI released Agnes-3.0-Flash Preview, an open-weights multimodal checkpoint with 33B parameters and a 262,144-token context window under Apache 2.0. The model supports text, image, and video understanding, tool calling, and adjustable reasoning effort. The repo clarifies this preview checkpoint is distinct from the production/API Agnes 3.0 Flash model, which uses a different configuration with a 1M-token context window. Reported reference results include IFBench 74.20 and SciCode 38.08 against peers such as Qwen3.6-35B-A3B, Kimi K2.5, and MiniMax M3.

Open-weight preview checkpoint: 33B parameters, 262,144-token context, Apache 2.0
Multimodal text, image, and video understanding with tool calling
Distinct from production/API checkpoint featuring 1M-token context
IFBench 74.20 and SciCode 38.08 compared with Qwen3.6-35B-A3B and Kimi K2.5

Read the original

Tagsagnes open-weights multimodal llm long-context apache-2.0 model-release

Full article1,466 words · extracted from huggingface.co · click to collapse

</p>

</p>

# Agnes-3.0-Flash Preview

## Model version clarification

This repository contains an earlier open-weight **Preview checkpoint** of Agnes 3.0 Flash. It is distinct from the newer **production/API checkpoint** listed on [Artificial Analysis](https://artificialanalysis.ai/models/agnes-3-0-flash).

The Preview release has **33B parameters** and a context window of **262,144 tokens**. The production/API model uses a different checkpoint and configuration, with a **1M-token context window**. Its benchmark results should not be attributed to the Preview weights released here.

This repository was initially published as `Agnes-3.0-Flash` without the `Preview` suffix. The model card now explicitly identifies this release as **Agnes-3.0-Flash Preview** to clarify the distinction between the open-weight release and the production/API model.

The specifications and Agnes benchmark results below refer to the **Preview checkpoint**.

Hello! 👋 Today we are introducing **Agnes-3.0-Flash Preview**, an **open-weights multimodal preview model** built for people who want flagship-class reasoning without flagship-class hardware.

Highlights:

- **Competitive across core capabilities.** Agnes-3.0-Flash Preview posts competitive results across reasoning, coding, and instruction-following evaluations.

- **Built for demanding work.** A **262 144-token context window**, adjustable reasoning effort, tool calling, and **text, image and video** understanding.

### Benchmarks

> **Benchmark scope:** The Agnes results in the chart and table below belong to the **Agnes-3.0-Flash Preview open-weight checkpoint** released in this repository. They are not results for the production/API Agnes 3.0 Flash model listed on Artificial Analysis.

</p>

The label "Agnes-3.0-Flash" in the benchmark image refers to this Preview checkpoint.

Reference results across contemporary models are shown below. The figures were compiled from different sources, harnesses, and model snapshots and do not constitute a controlled head-to-head comparison.

<th style="width:15%;padding:9px 6px;text-align:left;border-bottom:2px solid #3248AF;color:#3248AF;font-size:11px">Benchmark</th>

<th style="width:10%;padding:9px 5px;text-align:center;font-weight:700;border-bottom:2px solid #3248AF;color:#3248AF;background:rgba(50,72,175,.09);font-size:10px;line-height:1.3">Agnes-3.0-Flash Preview</th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Qwen3.6-35B-A3B<br><span style="opacity:.65">35B / 3B active</span></th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Kimi K2.5<br><span style="opacity:.65">1T / 32B active</span></th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Muse Glimmer<br><span style="opacity:.65">30B</span></th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">DeepSeek V4 Flash 0731<br><span style="opacity:.65">284B / 13B active</span></th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Gemini 3.5 Flash<br><span style="opacity:.65">undisclosed</span></th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">Qwen3.8 Flash Next<br><span style="opacity:.65">125B / 6B active</span></th>

<th style="width:8.33%;padding:9px 5px;text-align:center;border-bottom:2px solid #3248AF;font-size:10px;line-height:1.3">MiniMax M3<br><span style="opacity:.65">428B / 23B active</span></th>

</tr></thead><tbody>

<tr><td style="padding:7px 6px">IFBench</td><td style="padding:7px 5px;text-align:center;font-weight:700;color:#3248AF;background:rgba(50,72,175,.05)">74.20</td><td style="padding:7px 5px;text-align:center">64.4</td><td style="padding:7px 5px;text-align:center">43.7</td><td style="padding:7px 5px;text-align:center">77.0</td><td style="padding:7px 5px;text-align:center">75.6</td><td style="padding:7px 5px;text-align:center">75.8</td><td style="padding:7px 5px;text-align:center">79.5</td><td style="padding:7px 5px;text-align:center">76.3</td><td style="padding:7px 5px;text-align:center">81.3</td><td style="padding:7px 5px;text-align:center">82.9</td></tr>

<tr><td style="padding:7px 6px">SciCode</td><td style="padding:7px 5px;text-align:center;font-weight:700;color:#3248AF;background:rgba(50,72,175,.05)">38.08</td><td style="padding:7px 5px;text-align:center">35.8</td><td style="padding:7px 5px;text-align:center">39.6</td><td style="padding:7px 5px;text-align:center">43.6</td><td style="padding:7px 5px;text-align:center">39.5</td><td style="padding:7px 5px;text-align:center">50.3</td><td style="padding:7px 5px;text-align:center">46.6</td><td style="padding:7px 5px;text-align:center">53.1</td><td style="padding:7px 5px;text-align:center">50.6</td><td style="padding:7px 5px;text-align:center">45.4</td></tr>

<tr><td style="padding:7px 6px">GPQA Diamond</td><td style="padding:7px 5px;text-align:center;font-weight:700;color:#3248AF;background:rgba(50,72,175,.05)">85.05</td><td style="padding:7px 5px;text-align:center">84.1</td><td style="padding:7px 5px;text-align:center">78.9</td><td style="padding:7px 5px;text-align:center">83.5</td><td style="padding:7px 5px;text-align:center">85.8</td><td style="padding:7px 5px;text-align:center">90.8</td><td style="padding:7px 5px;text-align:center">90.5</td><td style="padding:7px 5px;text-align:center">92.2</td><td style="padding:7px 5px;text-align:center">92.3</td><td style="padding:7px 5px;text-align:center">92.9</td></tr>

<tr><td style="padding:7px 6px">AA-Omniscience Accuracy</td><td style="padding:7px 5px;text-align:center;font-weight:700;color:#3248AF;background:rgba(50,72,175,.05)">23.00</td><td style="padding:7px 5px;text-align:center">18.8</td><td style="padding:7px 5px;text-align:center">22.9</td><td style="padding:7px 5px;text-align:center">27.0</td><td style="padding:7px 5px;text-align:center">20.7</td><td style="padding:7px 5px;text-align:center">40.4</td><td style="padding:7px 5px;text-align:center">18.4</td><td style="padding:7px 5px;text-align:center">51.4</td><td style="padding:7px 5px;text-align:center">24.5</td><td style="padding:7px 5px;text-align:center">16.7</td></tr>

</tbody></table></div>

Higher is better for every row. Header parameter figures mix total and active counts, and harnesses and snapshot dates differ across sources, so treat cross-column comparisons as reference values rather than a controlled head-to-head evaluation.

</p>

## Architecture

Agnes-3.0-Flash Preview is a hybrid-attention decoder: three of every four layers run a gated delta rule (recurrent, with per-layer state independent of sequence length), and the fourth runs standard global attention. Only 18 of the 72 layers therefore hold a KV cache that grows with context.

| | |

|---|---|

| Context length | **262 144** tokens |

| Decoder layers | 72 = 54 delta-rule recurrent + 18 global attention, alternating 3 : 1 |

| Hidden size | 5120 |

| Global attention | 24 query heads / 4 KV heads (6 : 1 GQA), head dim 256; RMS-norm on q and k, sigmoid-gated output |

| Delta-rule layers | 16 key heads / 48 value heads, head dim 128; causal conv (kernel 4) in front, gated RMS-norm; recurrent state in fp32 |

| Feed-forward | SwiGLU, intermediate size 17408; plus a parallel SwiGLU 2048 branch in every layer |

| Positions | 3-axis rotary (text / height / width), interleaved mrope sections 11 : 11 : 10, base 1e7, applied to the first 25 % of each head dim (64 dims) |

| Vocabulary | 248 320 |

| Vision tower | 27 layers, hidden 1152, patch 16, 2 × 2 spatial merge, projected to 5120 |

## Quickstart

<div style="font-weight:700;color:#3248AF;margin-bottom:6px">REMOTE CODE REQUIRED</div>

<p style="margin:0"><b>Agnes-3.0-Flash Preview</b> ships its own model implementation. Always load it with <code>trust_remote_code=True</code>.</p>

</div>

### Requirements

```bash

pip install "transformers>=5.12" torch torchvision accelerate

```

Tested on transformers 5.12.1. Image and video inputs go through the bundled processor, which needs `torchvision`.

### Transformers

```python

from transformers import AutoModelForCausalLM, AutoTokenizer

path = "Agnes-AI/Agnes-3.0-Flash"

tok = AutoTokenizer.from_pretrained(path)

model = AutoModelForCausalLM.from_pretrained(

path, dtype="bfloat16", device_map="auto", trust_remote_code=True

)

msgs = [{"role": "user", "content": "请用三句话解释什么是人工智能。"}]

ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)

out = model.generate(ids, max_new_tokens=256)

print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))

```

### Images and video

Image and video inputs go through the bundled processor (also remote code):

```python

from transformers import AutoProcessor

proc = AutoProcessor.from_pretrained(path, trust_remote_code=True)

msgs = [{"role": "user", "content": [{"type": "image", "image": "photo.jpg"},

{"type": "text", "text": "描述这张图。"}]}]

inputs = proc.apply_chat_template(msgs, add_generation_prompt=True, tokenize=True,

return_dict=True, return_tensors="pt").to(model.device)

out = model.generate(**inputs, max_new_tokens=256)

print(proc.batch_decode(out[:, inputs["input_ids"].shape[1]:], skip_special_tokens=True)[0])

```

### Reasoning effort

The chat template exposes three reasoning levels — `high` (default), `medium`, `low` — plus a thinking-off switch:

```python

ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt",

reasoning_effort="medium") # or enable_thinking=False

```

### Tool calling

The chat template renders tool definitions for you. The model emits calls as `<tool_call><function=…><parameter=…>`, and you feed results back as a `tool` role message:

```python

tools = [{

"type": "function",

"function": {

"name": "get_weather",

"description": "Look up current weather for a city",

"parameters": {

"type": "object",

"properties": {"city": {"type": "string", "description": "City name"}},

"required": ["city"],

}]

msgs = [{"role": "user", "content": "What's the weather in Beijing right now?"}]

ids = tok.apply_chat_template(msgs, tools=tools, add_generation_prompt=True,

Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/Agnes-AI/Agnes-3.0-Flash