Apodex-1.1-mini-GGUF

Apodex-1.1-mini-GGUF

Reasoning model for long research tasks that works with files, data, code and tools, packaged as GGUF files you can host yourself. Suits researchers and developers who run their own models.

Made by apodex on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This is a GGUF version of Apodex 1.1 mini, a reasoning-first model for long, complex research tasks. It is based on Qwen3.5-35B-A3B and has 35.5 billion parameters. It is meant to take a task from raw inputs to a finished, checkable deliverable, using files, data and code along the way.

Who it suits: Researchers, analysts and developers who want a model they host themselves for multi-step work with documents and data.

Industry | Who uses it | How they use it | Example
Scientific research | Researchers | Hand over papers and datasets and ask for analysis with checked claims | A dataset cleaned, analysed and written up with sources
Finance | Analysts | Work through spreadsheets and financial questions in one long task | A model of a company built from several files
Software | Developers | Run the model behind an OpenAI-style server and connect their own tools | An agent that calls tools and shows its reasoning
Professional services | Consultants | Delegate parts of a long piece of work and add files as it goes | A report revised after new files arrive

Skip it if you need a small model for a basic laptop, since the weights are about 17.8 GB even at 4-bit.

Where to use it:

• Local machine: any Windows, macOS or Linux computer that has Python installed for fetching the files
• A self-hosted server started with the parsers for the model's chat format, so replies carry tool calls and reasoning in OpenAI-style fields
• The Agent Team code the Apodex team published, which the card suggests for trying this mini model

Use case | Who | What they get back
Long research task | Researchers | A deliverable built from papers, data and code, with claims checked
Tool-calling agent | Developers | Tool calls and reasoning returned in standard fields
Parallel subtasks | Teams | Several subagents sharing one task state and adjusting to feedback

How to set it up: You need Python with pip, a machine that can hold a 35.5B model, and about 17.8 GB of disk for the 4-bit size, which is an estimate.

• Python and pip on Windows, macOS or Linux
• Enough disk and memory for a model of this size
• A server that can load GGUF files

Install the Hugging Face hub tool and download the files.

pip install -U huggingface_hub
hf download apodex/Apodex-1.1-mini-GGUF

Start your own server on the downloaded GGUF file, using the parsers that match the Qwen3.5 chat template. Check that replies come back with standard tool_calls and reasoning_content fields.

How to use it: Send chat requests in the Qwen3.5 template. Tool calls and reasoning come out in their own tags, and the matching parsers turn them into standard fields. For the whole agent setup, the card points to the Agent Team code the maker published.

hf download apodex/Apodex-1.1-mini-GGUF

Tool | What it does
hf download | Fetches this repository's files to your computer
Agent Team | Splits a task between subagents that share one task state
Statement Review | Checks key claims against their sources, data and calculations before delivery

Example: You ask the model to clean a survey spreadsheet, compare two groups and write up the findings with sources. It works through the steps, can take new files while it runs, and checks the main claims against the data before handing the report back.

For newcomers and experts: If you are new to this, start with the download and a local server, then ask one question about a single file. Add tools once plain replies work.

If you know the field, the card describes an asynchronous Agent Team on top of AgentOS that keeps task state, adapts its plan and runs subagents in parallel. The weights follow the Qwen3.5 chat template. The trade is size, since a 35.5B model needs real hardware even when quantized.

You give a task and files
Agent Team splits the work
Subagents update a shared task state
Statement Review checks key claims
You get a checkable deliverable

Good to know: The card reports its own benchmark scores for the mini model, so treat them as the maker's figures.

Title: Apodex-1.1-mini benchmark scores
FrontierFinance: 50.2
APEX-Agents: 27.7

• Size: about 17.8 GB at 4-bit, an estimate, in GGUF format
• Languages: English and Chinese
• License: Apache 2.0
• Cost to run: the files are free to download. The card also names an online service and an API from the maker, with no prices given
• Data it reaches: the files, data and code you give it, plus any tools you connect
• Not for: none named

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 35.5B
• Weights: about 17.8 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 479
• Likes: 22
• License: apache-2.0
• Languages: en, zh
• Based on: Qwen/Qwen3.5-35B-A3B
• Tags: transformers, gguf, agent, apodex, text-generation, conversational

Links

• Model: https://huggingface.co/apodex/Apodex-1.1-mini-GGUF
• Files: https://huggingface.co/apodex/Apodex-1.1-mini-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download apodex/Apodex-1.1-mini-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.