Instructions to use batiai/Mistral-Medium-3.5-128B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use batiai/Mistral-Medium-3.5-128B-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS
# Run inference directly in the terminal:
llama cli -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS
# Run inference directly in the terminal:
llama cli -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS
# Run inference directly in the terminal:
./llama-cli -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS
# Run inference directly in the terminal:
./build/bin/llama-cli -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Use Docker

docker model run hf.co/batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

LM Studio
Jan

vLLM

How to use batiai/Mistral-Medium-3.5-128B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "batiai/Mistral-Medium-3.5-128B-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "batiai/Mistral-Medium-3.5-128B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Ollama
How to use batiai/Mistral-Medium-3.5-128B-GGUF with Ollama:
```
ollama run hf.co/batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS
```
Unsloth Desktop

How to use batiai/Mistral-Medium-3.5-128B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use batiai/Mistral-Medium-3.5-128B-GGUF with Docker Model Runner:
```
docker model run hf.co/batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS
```

Lemonade

How to use batiai/Mistral-Medium-3.5-128B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Run and chat with the model

lemonade run user.Mistral-Medium-3.5-128B-GGUF-IQ3_XXS

List all available models

lemonade list

Hermes Agent

How to use batiai/Mistral-Medium-3.5-128B-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use batiai/Mistral-Medium-3.5-128B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "batiai/Mistral-Medium-3.5-128B-GGUF:IQ3_XXS" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Mistral Medium 3.5 128B GGUF — Quantized by BatiAI

Quantizations of Mistral Medium 3.5 128B (Dense, Modified MIT) — frontier coding model on a Mac. Built and verified by BatiAI for BatiFlow.

Why Mistral Medium 3.5?

SWE-Bench Verified: 77.6% — second only to Gemini 3.1 Pro Preview (78.8%)
Tau3-Telecom: 91.4% — top-tier agentic benchmark
128B Dense — Mistral's first flagship merged model (instruction + reasoning + coding in one weight set)
Modified MIT — open weights, commercial-friendly
256K context window + multimodal vision capabilities
Configurable reasoning effort per request
Released April 29, 2026

Quick Start

ollama pull batiai/mistral-medium-3.5:iq4

Available Quantizations

Quant	Size	VRAM target	Recommended For
IQ3_XXS	~50GB	~64GB	96GB+ Mac (compact)
IQ4_XS	~70GB	~84GB	128GB Mac (recommended)
Q5_K_M	~90GB	~108GB	192GB+ Mac (highest quality)

RAM Requirements

Your Mac RAM	IQ3 (~50GB)	IQ4 (~70GB)	Q5 (~90GB)
64GB	⚠️ Heavy swap	❌	❌
96GB	✅ Fast	⚠️ Tight	❌
128GB	✅	✅ Recommended	⚠️ Tight
192GB+	✅	✅	✅ Best
512GB	✅	✅	✅ Headroom

Coding Powerhouse on a Mac

Benchmark	Mistral Medium 3.5
SWE-Bench Verified	77.6%
Tau3-Telecom (agentic)	91.4%
Context window	256K

A coding model that goes head-to-head with cloud frontier (Gemini 3.1 Pro), runnable entirely on-device on a 128GB MacBook Pro.

Model Comparison — Frontier Class on Mac

Your Mac	Best Pick	Speed	Use Case
32GB	`batiai/nemotron3-nano:iq4`	—	General agentic
96GB	`batiai/mistral-medium-3.5:iq3` (this)	—	Coding agents
128GB	`batiai/mistral-medium-3.5:iq4` (this)	—	Frontier coding
128GB	`batiai/minimax-m2.7:iq3`	36.7 t/s	229B Dense, GDPval
192GB+	`batiai/mistral-medium-3.5:q5`	—	Max quality
512GB	`batiai/kimi-k2.6:iq4`	—	1T MoE, SWE-Bench Pro 58.6

Why BatiAI Quantization?

	BatiAI	Third-party
Source	Official Mistral weights	Re-quantized
imatrix	✅ wikitext-2 200 chunks	Standard
Vision	(text-only quants)	(text-only quants)
BatiAI signed	✅	—

Technical Details

Original Model: mistralai/Mistral-Medium-3.5-128B
Architecture: Dense decoder-only, 128B params
License: Modified MIT (commercial-friendly)
Quantized with: llama.cpp + imatrix
Quantized by: BatiAI

Config fix applied (YaRN, 2026-06-16)

The upstream config bug (mscale_all_dim 1→0, fixed here) propagated to the GGUF metadata as mistral3.rope.scaling.yarn_log_multiplier = 1.0 (correct value 0.0). Since YaRN scaling is read from metadata at load time — not baked into weights — we patched the value in place; no re-quantization needed. All quants (IQ3_XXS / IQ4_XS / Q5_K_M) are patched and live (verified 0.0). Filenames unchanged — just re-download to pick up the fix. For long-context workloads, vLLM is the recommended runtime.

About BatiFlow

BatiFlow — free, on-device AI automation for Mac. 5MB app, 100% local, unlimited. 60+ tools — KakaoTalk, iMessage, Slack, Calendar, browser, file system.