Mistral Large 4 (1T-A52B) Architecture Parameters, Benchmarks & VRAM
Mistral flagship 1T-parameter (52B active) native multimodal MoE with 1M context and 262K max output.
🧬 Mistral Large 4 (1T-A52B) Architecture & Tensor Specifications
Verified directly against HuggingFace (mistralai/Mistral-Large-4-1T-A52B) config.json & Model Card
1. Backbone & Dimensions
mistral_large4- Total / Active Params
- 1T / 52B
- Hidden Size (d_model)
- 6,144
- FFN Intermediate Size
- 24,576
- MoE Expert Dim
- 6,144
- MoE Experts
- 128 Routed · Top-4 + 1 Shared
- Activation / Norm
- SiLU (SwiGLU) · ε=0.00001
- Native Tensor Dtype
- BF16 / FP8
2. Attention & Layer Layout
MLA- Num Hidden Layers
- 72 Layers
- Hidden Layer Layout
- 72-Layer Sparse MoE (128 Experts, Top-4 Active)
- Attention Heads (Q / KV)
- 64 Q / 64 KV
- Head Dimension
- 64
- Vision Encoder / MTP
- ViT 30L
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 1,048,576 tokens
- RoPE Configuration
- RoPE (原生 1,048,576 Context)
- Vocab Size
- 204,800 (Untied)
📋 Mistral Large 4 (1T-A52B) Official API & System Card Specifications
Verified against official Mistral AI developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- mistral-large-4
- Release Date
- 2026-10-06
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- ✅ Supported
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 1,024,000 tokens (1024K)
- Max Output Tokens
- 262,000 tokens
- Input Modalities
- Text + Image + Video
- Output Modalities
- Text
- Direct China Access
- 🌐 Global / OpenRouter
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
USD / 1M- Input / 1M Tokens
- $ 0.7
- Cached Input / 1M
- $ 0.07 -90%
- Output / 1M Tokens
- $ 2.1
- Tiered Pricing Notes
- Mistral AI: Mistral La Plateforme API (includes free experiment tier)
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
📊 Mistral Large 4 (1T-A52B) Full Benchmark Scorecard
14 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 7 items
CyBench is a suite of Capture-the-Flag (CTF) challenges measuring agentic cyber attack capabilities. It evaluates dual-use cybersecurity knowledge and measures the 'unguided success rate', where agents complete tasks end-to-end without guidance on appropriate subtasks.
SciCode is a research coding benchmark curated by scientists that challenges language models to code solutions for scientific problems.
CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and complete security-related challenges in a controlled environment.
DeepSWE 1.1 evaluates software engineering agents using the mini-swe-agent harness.
SWE Atlas - Codebase QnA evaluates a model's ability to answer questions about real codebases, measuring repository-level comprehension and the ability to reason about code structure, behavior, and intent across an entire project.
Terminal-Bench 4.0 evaluates AI agents on real-world work in terminal and command-line containerized environments across 66 tasks, with emphasis on science-adjacent and frontier engineering problems.
FrontierCode 2.0 is Cognition's private benchmark assessing end-to-end code quality of coding-agent changes, including whether a repository maintainer would merge the agent's change.
🤖 Agents & Tools 5 items
BrowseComp is a benchmark comprising 1,266 questions that challenge AI agents to persistently navigate the internet in search of hard-to-find, entangled information.
AutomationBench is a tool-use benchmark that evaluates AI agents on automating real-world workflows, testing their ability to orchestrate tools and complete multi-step automation tasks.
Finance Agent v2 is an agentic financial-analysis benchmark from Vals that evaluates models on real-world finance workflows, measuring their ability to retrieve and reason over financial documents, perform multi-step calculations, and produce accurate analyses.
AA-Briefcase is an Artificial Analysis evaluation of AI systems on professional knowledge-work tasks, reported as an Elo score.
Harvey's Legal Agent Benchmark evaluates AI agents on complex legal workflows and reports the all-pass rate.
👁️ Multimodal 2 items
ChartQAPro is a challenging benchmark for question answering over diverse, real-world charts and infographics.
GDP.pdf is a knowledge-work vision benchmark that evaluates models on economically valuable professional tasks presented as visual documents (PDFs), testing document-based reasoning, chart and table interpretation, and problem solving without tools.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ Mistral Large 4 (1T-A52B) Interactive VRAM & Token Speed Calculator
Physical 96 GB · Load 660.1%
Exceeds usable VRAM by 526.95 GB
MLA compresses multi-batch KV footprint
OLLAMA_CONTEXT_LENGTH=16384 ollama run mistral-large-4-1tInsufficient Local VRAM? Ways to Run Mistral Large 4 (1T-A52B)
Rent an hourly high-VRAM cloud GPU pod, or check the 12-month Buy vs Cloud vs API break-even curve.
📊 Mistral Large 4 (1T-A52B) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 2000 GB | 1.36 GB | 2042.66 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 1062.5 GB | 1.36 GB | 1086.41 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 820 GB | 1.36 GB | 839.06 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 711.25 GB | 1.36 GB | 728.13 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 606.25 GB | 1.36 GB | 621.03 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 531.25 GB | 1.36 GB | 544.53 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 488.75 GB | 1.36 GB | 501.18 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 393.75 GB | 1.36 GB | 404.28 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for Mistral Large 4 (1T-A52B) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of Mistral Large 4 (1T-A52B)?
Mistral Large 4 (1T-A52B) uses MistralLarge4ForConditionalGeneration with 72 layers (72-Layer Sparse MoE (128 Experts, Top-4 Active)), hidden_size=6144, 64 Q heads / 64 KV heads (head_dim=64), and vocab_size=204,800.
How much VRAM does Mistral Large 4 (1T-A52B) require?
At 4-bit quantization (Q4_K_M, 16K context), Mistral Large 4 (1T-A52B) requires 621.03 GB VRAM. Full FP16 requires 2042.66 GB.