Leanstral 1.5 119B A6B — Hardware Requirements & GPU Compatibility

ChatFunctions

Leanstral 1.5 119B A6B is Mistral AI's open-source code agent model specialized for Lean 4, the proof assistant used to formalize complex mathematics and software specifications. Built as part of the Mistral Small 4 family, it is a mixture-of-experts model with 128 experts and 4 active per token, totaling 119 billion parameters with about 6.5 billion active per token, and it accepts both text and image input while producing text output. It is designed to work through long formal-proof tasks over many hours via the Mistral Vibe CLI and the lean-lsp-mcp tool, with a "high" reasoning-effort mode for complex proofs. Because all 128 experts must be held in memory regardless of how few activate per token, it still needs a multi-GPU setup to run locally. Context length is 256,000 tokens, though Mistral recommends staying under 200,000 tokens in practice. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in July 2026 as an update to the original Leanstral model.

119B params · MoE6B activeApache 2.0Jul 2026393 downloads225 likes2.1K quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Leanstral 1.5 119B A6B runs on a 96 GB+ Mac or a workstation GPU

Best local pick → Mac Studio (M3 Ultra, 96GB) · Q4_K_M · 84 GB · ~61 tok/s

Macs from 96 GB (Q4_K_M · ~61 tok/s)

Apache 2.0

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Leanstral 1.5 119B A6B

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on H200 (141 GB) (57 GB headroom) · Q4_K_M

Generation speed: ~256 tok/s; generation speed
Cost per 1M output tokens: $3.90; per 1M output tokens

Deployon RunPodBest optionfrom $3.59/hr

Compare GPUs →

Run Leanstral 1.5 119B A6B

84.0 GB VRAM at 16K context · runs on one RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1. Splits the model over 8 GPUs of one server (tensor parallel).
$vllm serve mistralai/Leanstral-1.5-119B-A6B --max-model-len 16384 --tensor-parallel-size 8
Runs the original weights, not the Q4_K_M GGUF; needs ~262 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model mvid/Leanstral-1.5-119B-A6B-MLX-4bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model mvid/Leanstral-1.5-119B-A6B-MLX-4bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$GZGavinZhao/Leanstral-1.5-119B-A6B-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
H200 SXM (141 GB)CheapestCheapest57 GB free	~256 tok/s	$3.90	Deploy on RunPodfrom $3.59/hr (opens in new tab)	141 GB · 57 GB free · ~256 tok/s · $3.90 / 1M tokens
H200 NVL (141 GB)57 GB free	~256 tok/s	$4.61	Deploy on Vast.aifrom $4.24/hr (opens in new tab)	141 GB · 57 GB free · ~256 tok/s · $4.61 / 1M tokens
B200 (192 GB)108 GB free	~294 tok/s	$5.65	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 108 GB free · ~294 tok/s · $5.65 / 1M tokens
B300 (288 GB)204 GB free	~294 tok/s	$6.56	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 204 GB free · ~294 tok/s · $6.56 / 1M tokens

Compare GPUs →

How Much VRAM Does Leanstral 1.5 119B A6B Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	Runs on
55.6 GB	67.3 GB	104.8 GB	~50.6 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Runs on Instinct MI210 64 GB
63.8 GB	75.5 GB	113 GB	~58 GB	3-bit medium — 3-bit medium quantization	Runs on RTX PRO 5000 Blackwell 72 GB
78.5 GB	90.3 GB	127.7 GB	72.2 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
93.3 GB	105 GB	142.5 GB	~84.8 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Runs on Instinct MI250X 128 GB
108.0 GB	119.7 GB	157.2 GB	97.7 GB	6-bit, very good — 6-bit quantization, very good quality	Runs on Instinct MI250X 128 GB
130.9 GB	142.6 GB	180.1 GB	~119 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Runs on H200 NVL 141 GB
261.8 GB	273.5 GB	311 GB	238 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Runs on Instinct MI350X 288 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

84.0GBat 16K context

Weights: 77.5 GB
KV cache: 6.2 GB
Overhead: 0.3 GB

Server-class: ~84 GB at 16K context · one RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB

Or rent H200 from $3.59/hr.

Which Hardware Can Run Leanstral 1.5 119B A6B?

Q4_K_M · 78.5 GB

Leanstral 1.5 119B A6B (Q4_K_M) requires 78.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 103+ GB is recommended. No consumer GPU has enough memory. 30 devices with unified memory can run Leanstral 1.5 119B A6B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 61.1 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

Alternatives to Leanstral 1.5 119B A6B

Mistral family · 34 models →

Similar models

Model	Score	VRAM	tok/s	Why it is similar
GPT OSS 120BOpenAI	97.5	70.8 GB	—needs > 24 GB	same size
GLM 4.5 AirZ.ai	96.3	67.6 GB	—needs > 24 GB	same size
C4ai Command A 03 2025Cohere	79.8	78.7 GB	—needs > 24 GB	same size
NVIDIA Nemotron 3.5 Super VL 120B A12B BF16NVIDIA	—	87.7 GB	—needs > 24 GB	same size
Ling 3.0 Flash VLInclusion AI	—	76 GB	—needs > 24 GB	same size
Qwen1.5 110BAlibaba	—	72.4 GB	—needs > 24 GB	same size

Specifications

Publisher: Mistral AI
Family: Mistral
Parameters: 119B
Active parameters: 6B per token
Release Date: 2026-07-01
License: Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Leanstral 1.5 119B A6B need?

Leanstral 1.5 119B A6B requires 78.5 GB of VRAM at Q4_K_M, or 261.8 GB at BF16.

VRAM = Weights + KV Cache + Overhead

Weights = 119B × 4.8 bits ÷ 8 = 71.4 GB

KV Cache + Overhead ≈ 7.1 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

VRAM usage by quantization

Q4_K_M

78.5 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Leanstral 1.5 119B A6B?

No — Leanstral 1.5 119B A6B requires at least 55.6 GB at Q2_K, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Leanstral 1.5 119B A6B?

For Leanstral 1.5 119B A6B, Q4_K_M (78.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (93.3 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 55.6 GB.

VRAM requirement by quantization

Q2_K

55.6 GB~75%

Q4_K_M ★

78.5 GB~89%

Q5_K_M

93.3 GB~92%

Q6_K

108.0 GB~95%

Q8_0

130.9 GB~99%

BF16

261.8 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Leanstral 1.5 119B A6B on a Mac?

Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run Leanstral 1.5 119B A6B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 78.5 GB of usable unified memory (RAM minus macOS overhead).

Can I run Leanstral 1.5 119B A6B locally?

Yes — Leanstral 1.5 119B A6B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Leanstral 1.5 119B A6B?

At Q4_K_M, Leanstral 1.5 119B A6B can reach ~96 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 4.0 GB active ÷ (8000 × 0.65) = 0.76 ms, plus 48 layers × 0.055 ms = 2.64 ms, so 1000 ÷ 3.40 ms = ~294 tok/s

Estimated speed at Q4_K_M (78.5 GB)

NVIDIA B200

~294 tok/s

NVIDIA B300

~294 tok/s

NVIDIA GH200 Grace Hopper Superchip

~258 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Leanstral 1.5 119B A6B?

At Q4_K_M, the download is about 71.40 GB. The full-precision BF16 version is 238.00 GB. The smallest option (Q2_K) is 50.58 GB.

Which GPUs can run Leanstral 1.5 119B A6B?

No single consumer GPU has enough VRAM to run Leanstral 1.5 119B A6B at Q4_K_M (78.5 GB). Multi-GPU or professional hardware is required.

Which devices can run Leanstral 1.5 119B A6B?

30 devices with unified memory can run Leanstral 1.5 119B A6B at Q4_K_M (78.5 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.