What is the maximum context window of Mistral Large 4 "Le Chonk" (1T MoE)?

Mistral Large 4 "Le Chonk" (1T MoE) supports a maximum context window of 1,048,576 tokens (1M) (1,048,576 tokens) and up to 131,072 output tokens using Granular MoE (~1,050B Total / 52B Active, 128 routed + 1 shared expert) + MLA + DSA sparse attention + 1.6B Pixtral-style vision encoder.

How much GPU VRAM is required to run Mistral Large 4 "Le Chonk" (1T MoE) locally?

At 4-bit quantization (Q4_K_M / AWQ) with a 16K context window, Mistral Large 4 "Le Chonk" (1T MoE) (~1,050B Total / 52B Active) requires approximately 632.71 GB of total VRAM (630 GB for weights + 2.7 GB for KV-cache and runtime). In full 16-bit precision (FP16/BF16), it requires approximately 2103.27 GB of VRAM. Recommended hardware: 8× H200 / B200 Node (FP8) or 8× B200 (NVFP4 checkpoint).

Mistral AI OPEN-WEIGHT · Mistral Ecosystem Open-Weight License (modified MIT; MaaS needs commercial license) · Context: 1,048,576 tokens (1M)

Mistral Large 4 "Le Chonk" (1T MoE)

Name: Mistral Large 4 "Le Chonk" (1T MoE)
Price: 0.7 USD
Author: Mistral AI

Mistral's largest model and first trillion-parameter open-weight release: a European-trained multimodal MoE with 52B active parameters, a 1M-token context, and three reasoning efforts, claiming the open-weight lead on FrontierCode 2.0 and Terminal-Bench 4.0.

Provider

Mistral AI (Mistral Ecosystem Open-Weight License (modified MIT; MaaS needs commercial license))

Architecture

Granular MoE (~1,050B Total / 52B Active, 128 routed + 1 shared expert) + MLA + DSA sparse attention + 1.6B Pixtral-style vision encoder

Parameters

~1,050B Total / 52B Active

Context Window

1,048,576 tokens (1M) (Max Output: 131,072 tokens)

API Pricing

Input: $0.70 / 1M · Cached: $0.07 / 1M · Output: $2.10 / 1M

VRAM (16K Ctx)

Q4_K_M: 632.71 GB · Q8_0: 1115.77 GB · FP16: 2103.27 GB

Hardware Target

8× H200 / B200 Node (FP8) or 8× B200 (NVFP4 checkpoint)

Interactive Token Cost Estimator (Mistral Large 4 (1T))

Requests / Day

Input Tokens / Req

Output Tokens / Req

GPU Hardware Compatibility Matrix (Mistral Large 4 (1T) at 16K Context)

GPU / Hardware Profile	VRAM	Q4_K_M (4.8-bit)	Q8_0 (8-bit)	Est. Speed (Q4)
RTX 5060 Ti 16GB 448 GB/s	16 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~2 tok/s
RTX 4090 24GB 1008 GB/s	24 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~5 tok/s
RTX 5090 32GB 1792 GB/s	32 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~8 tok/s
M4 Max 64GB 546 GB/s	52 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~2 tok/s
2× 24GB (48GB) 920 GB/s	48 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~4 tok/s
H100 80GB 3350 GB/s	80 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~15 tok/s
M4 Ultra 192GB 819 GB/s	164 GB	REQUIRES MULTI-GPU (632.71 GB)	REQUIRES MULTI-GPU (1115.77 GB)	~4 tok/s
8× H200 (1,128GB) 4800 GB/s	1128 GB	FITS IN VRAM (632.71 GB)	TIGHT FIT (Q8 KV) (1115.77 GB)	~101 tok/s

Architectural Strengths & Target Workloads

Vendor-reported open-weight lead on FrontierCode 2.0 (19.7%) and Terminal-Bench 4.0 (41%) with high token efficiency
51% on the Artificial Analysis Cyber Index with a 96.6% refusal rate on malicious cyber prompts
Low / medium / high reasoning efforts per request, native strict JSON schema output, and an embedded DSpark speculative decoding head

Primary Engineering Workloads: Agentic Coding at Scale · Defensive Security & Vulnerability Triage · Sovereign / On-Prem Enterprise Deployments.

Frequently Asked Questions (Mistral Large 4 (1T))

How much does Mistral Large 4 "Le Chonk" (1T MoE) cost per 1M tokens?

Mistral Large 4 "Le Chonk" (1T MoE) is priced at $0.70 per 1M uncached input tokens, $0.07 per 1M cached input tokens, and $2.10 per 1M output tokens (Mistral API (mistral-large-4) / OpenRouter / Microsoft Foundry / NVIDIA NIM, or Self-Hosted vLLM / SGLang).

What is the context window of Mistral Large 4 "Le Chonk" (1T MoE)?

It supports up to 1,048,576 tokens (1M) (1,048,576 tokens) and a maximum generation output of 131,072 tokens.

Full TechPulse Engineering Analysis

Mistral Opens the Weights of Large 4 'Le Chonk': A 1T-Parameter MoE That Leads Open Models on Coding and Cyber Defense

Three weeks after its preview, Mistral's European-trained flagship ships on Hugging Face with a 1M-token context, three reasoning modes, and API pricing of $0.70 / $2.10 per million tokens.

Read Full Technical Dispatch (5 min read) →