Mistral Open-Sources 1T-Parameter Le Chonk

Episode 1 · Mistral Open-Sources Trillion-Parameter Flagship Mistral Large 4 "Le Chonk" (2026-10-29, 37 posts)

On October 29, Mistral updated and open-sourced the weights of its flagship model Mistral Large 4 (codename Le Chonk). It is Mistral's largest and most powerful model to date, and another major open-source move by a leading European lab.

Confirmed

• The model is a natively multimodal MoE with roughly 1.05T total parameters (some reports describe it as trillion-parameter scale), activating only about 49–52B parameters per token
• It scores 26 (high reasoning) on the Artificial Analysis Intelligence Index v5, up from the Preview version's 25, edging out the previous leading non-Chinese open-source model by about 1 point
• The weights are available on Hugging Face, with the official tagline "#1 in Sovereign AI"
• International outlets including The Next Web also covered the weight release

Why it matters

• Open weights make Mistral Large 4 the strongest non-Chinese open-source model, reshaping the competitive landscape of the open-source camp
• With its massive parameter scale combined with open weights, local/private deployment communities can fine-tune and deploy this frontier-grade model themselves—fitting neatly into Europe's "sovereign AI" narrative

• Unsloth ships GGUF quantized Mistral-Large-4-1T-A52B for local inference — unsloth · 2026-10-29
• Mistral Large 4 'Le Chonk' Released as Open Weights — Few_Painter_5588 · 2026-10-29
• Mistral releases Mistral Large 4: open-weight 1T-param MoE with 52B active, trained on 3800 Blackwell GPUs — jacek2023 · 2026-10-29
• Mistral open-sources Le Chonk, tops FrontierCode among open models for agentic coding — MistralAI · 2026-10-29
• Mistral open-weights 1T-parameter Mistral Large 4 'Le Chonk', tops FrontierCode — MistralAI · 2026-10-29
• Mistral Large 4 'Le Chonk' lands on Hugging Face: 1T-param MoE, 49B active, 1M context — ariG23498 · 2026-10-29
• Mistral's 1T-parameter Le Chonk lands on Hugging Face with llama.cpp support — Disastrous-Work-1632 · 2026-10-29
• Mistral releases open weights for its 1-trillion-parameter Large 4 model — joe4942 · 2026-10-29
• Mistral Large 4 GGUF appears on HF: 1T total params, 52B active — ngxson · 2026-10-29
• Mistral-Large-4-1T-A52B GGUF appears on Hugging Face: 1T-param MoE flagship leak — ngxson · 2026-10-29
• Mistral open-sources 1T-parameter MoE model Le Chonk, trained on 3,800 Grace Blackwell GPUs — Bam4d · 2026-10-29
• Mistral Large 4 goes open weights: 1.05T params, tops non-China open models on AA Index — ArtificialAnlys · 2026-10-29
• Mistral Large 4 goes open weights: 1.05T params, 52B active, top non-China open model — ArtificialAnlys · 2026-10-29
• Mistral Large 4: 1T-parameter open-weight multimodal model tops coding benchmarks — petrusenko_max · 2026-10-29
• Mistral releases Mistral Large 4 weights on Hugging Face, codenamed Le Chonk — _AndrewZhao · 2026-10-29
• Mistral Large 4 Scorecard: 49.8% Coding Agent Index, 93.3% Attack Resistance on Lakera B3 — _AndrewZhao · 2026-10-29
• Mistral Releases Open Weights of 1T-Parameter Large 4, Tops FrontierCode Among Open Models — _AndrewZhao · 2026-10-29
• Mistral Large 4 goes open weight: 1T-param MoE, Terminal-Bench up 13 points — _AndrewZhao · 2026-10-29
• Mistral Large 4, a 1T-parameter multimodal MoE, trends on Hugging Face — mistralai · 2026-10-29
• Mistral open-sources 1T-parameter MoE with 1M context, tops FrontierCode among open models — _AndrewZhao · 2026-10-29

17 more related posts →

Episode 2 · Unsloth and Hugging Face Bring Le Chonk Quantized for Local Deployment (2026-10-29, 3 posts)

Quantized versions of Mistral's Le Chonk are now available: Unsloth's 1bit version compresses 80% while retaining 91% quality and runs on a Mac Studio, while Hugging Face ships FP8, NVFP4 and GGUF formats for local deployment.

• Le Chonk Goes Local: FP8, NVFP4 and GGUF Quants Released — pcuenq · 2026-10-29
• Le Chonk goes local: open release ships FP8, NVFP4 and GGUF quants — pcuenq · 2026-10-29
• Unsloth quants run 1T-param Mistral Large 4 on Mac Studio: 1-bit keeps 91% accuracy at 223GB — danielhanchen · 2026-10-29

Episode 3 · Mistral Large 4 Benchmarks Show Top Token Efficiency, Second Among Open Coding Agents (2026-10-29, 4 posts)

Artificial Analysis's full evaluation of Mistral Large 4 (High) shows an Intelligence Index of 26 (rank 8) at 161 tokens/s, with class-leading efficiency of 82K output tokens per task (39% fewer than Preview); it scored 48 on the Coding Agent Index, second among open-weight models while using about half the tokens.

• Mistral Large 4 ranks #2 open-weights on Coding Agent Index using half the tokens — ArtificialAnlys · 2026-10-29
• Mistral Large 4 cuts output tokens 39% vs Preview, topping the open-weights Pareto line — ArtificialAnlys · 2026-10-29
• Mistral Large 4 (high) hits open-weights Pareto frontier at 82K output tokens per task — ArtificialAnlys · 2026-10-29
• Mistral Large 4 (High) scores 26 on Intelligence Index at 161 tokens/s — ArtificialAnlys · 2026-10-29