Apex-35B-V1.0

Apex-35B-V1.0

A text model for coding, tool use, question answering, search and research tasks, built on Qwen3.6-35B-A3B with about 3B parameters used per token. For developers and researchers.

Made by Apexintelligence-AI 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: Apex-35B-V1.0 is the debut release from Apex Intelligence. It starts from the Qwen3.6-35B-A3B model and gets extra training on private data and research trajectories from the company. It has 35 billion parameters in total, and about 3 billion are used for each token.

The makers pair deep search, question answering, tool use and coding with research tasks in computational science and robotics simulation. The card says it works with Claude Code.

Who it suits: Developers who want a coding and agent model they can run themselves, and researchers who want a model that plans computations and uses the results.

Industry | Who uses it | How they use it | Example
Software | Developers using coding agents | Run a local model behind an agent | Ask it to work on a code repository
Research | Computational scientists | Let the model plan and run programs | Revise an approach from execution feedback
Robotics | Simulation engineers | Use it for simulation tasks | Work through a robotics simulation problem

Skip it if you need image input or a small download, since this is a text model and the full weights are large.

Where to use it:

• Claude Code, which the card says supports this model
• Local runtimes such as llama.cpp and Unsloth, through the separate GGUF repository
• Python with Transformers, for the Safetensors files
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Coding | Developers | Code written and fixed across a repository
Deep search | Analysts | Answers built from search and tool use
Research computing | Scientists | Programs written and run, with the approach revised from the results

How to set it up:

• Python and pip to run the download commands
• About 71.9 GB of space for the BF16 weights, or about 18 GB at 4-bit by Hugging Face estimate
• A runtime that can load the files, which the model card describes

1. Install the Hugging Face tools.
2. Download the model files.
3. Load them in your runtime, or pick a GGUF build for a local runner.

pip install -U huggingface_hub
hf download Apexintelligence-AI/Apex-35B-V1.0

How to use it: Load the model in your runtime and chat with it, or connect it to a coding agent. It decides on each turn whether to reason and for how long, with no mode switch.

Property | Apex-35B-V1.0
Total parameters | 35B
Active per token | About 3B
Context window | 262,144 tokens, used for the prompt and the reply together
GGUF builds | BF16 plus 28 quantizations in a separate repository

Title: Size of the weights in GB
BF16 weights: 71.9 GB
Largest GGUF, Q8_0: 37.8 GB
4-bit estimate: 18 GB
Smallest GGUF, IQ1_S: 9.0 GB

Example: You connect the model to a coding agent and point it at a repository. It reasons when the task needs it, calls tools, and revises its work as results come back.

For newcomers and experts: If you are new to this, start with a GGUF build in a local runner so the download is smaller, and try a short coding question first.

If you know the field, it is a sparse mixture-of-experts model, with the long context shared between prompt and reply. The makers describe an automated research system that helped build training data, run experiments and tune GPU kernels.

Download the model or a GGUF build
Load it in your runtime
Send a coding or research task
Model reasons, uses tools and revises

Good to know:

• The makers say recursive self-improvement through Apex Research fed into the training data and recipe
• Hugging Face lists 12 quantized versions made by other people and none fine-tuned
• The card's evaluation charts are not reproduced here, so check them on the model page
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download and the prompts, code and tool results you give it
• Not for: image input

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 36B
• Weights: about 71.9 GB in BF16; about 18 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,987
• Likes: 75
• License: apache-2.0
• Based on: Qwen/Qwen3.6-35B-A3B
• Quantized versions: 12
• Tags: transformers, safetensors, qwen3_5_moe, image-text-to-text, mixture-of-experts, agent, research, text-generation, conversational

Links

• Model: https://huggingface.co/Apexintelligence-AI/Apex-35B-V1.0
• Files: https://huggingface.co/Apexintelligence-AI/Apex-35B-V1.0/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Apexintelligence-AI/Apex-35B-V1.0

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