Mistral releases Mistral Large 4, a 1T-parameter open-weight model
10/06/2026 — 10/08, 12:17·13 sources (all)·20 reports (all)·In the 2026-10-07 briefing·In the 2026-10-06 briefing
Key takeaways
- Mistral released Mistral Large 4 (codename Le Chonk), a 1T-parameter open-weight model with 49B active parameters and native multimodality,…
- Mistral claims SOTA on security-critical workloads.
Background & analysis
On October 6, 2026, French AI lab Mistral AI released a public preview of Mistral Large 4, known unofficially as ML4 and nicknamed le Chonk. Developers can access it through Mistral's API in Mistral Studio starting that day, while the model weights are scheduled to be released for download at the end of October. Mistral describes it as the most powerful open-weight model outside the US and China, and says it reaches SOTA among open models on enterprise workloads such as cybersecurity, finance and legal work. According to the company blog, the model was trained in Mistral's own data centers in Europe, with a large share of its training data being multilingual.
Reports differ on the specifications. Most describe a 1T-parameter model with 49B active parameters, while Unite.AI and MarkTechPost list 1.05T total parameters with 49B active; a Chinese report gives 52B active parameters. The GPU count also varies: Techmeme, citing The Deep View, says 4,000 Grace Blackwell GPUs, whereas IT之家, MarkTechPost and the official blog say 3,800. The model is described as a natively multimodal mixture-of-experts system, and MarkTechPost notes a 1 million token context window.
On benchmarks, Artificial Analysis reports a score of 38 on its Intelligence Index, calling Mistral Large 4 the most intelligent model from outside the US and China and comparable to DeepSeek V4.1 Flash (max). By country, France moves to third place, behind the US (Claude Opus 5.5 at 58) and China (MiMo-V2.6-Pro at 46). On the Cyber Index the model scores 50, level with GLM-5.3-Flash and behind MiMo-V2.6-Pro at 56. TestingCatalog reports 61.7% on DeepSWE v1.1, 93% on Cybench, and 42% on Dense 200 visual grounding, slightly above GPT-6 Astra at 41%. The Decoder notes a large gain on the Intelligence Index but says the model still trails Claude, GPT-6 and Chinese competitors.
Artificial Analysis lists standard pricing at $1.36 per 1M input tokens and $4.18 per 1M output tokens, with $0.14 per 1M cached input tokens, served at a 50% launch discount for the first two weeks. IT之家 reports that Mistral is running red-team testing with cybersecurity organizations, partners and regulators in real business environments, and that the test version has somewhat relaxed content safety policies. The rollout currently stands at this public preview, with weights not yet released.
AI-generated from 20 reports · updated 34 days ago
Latest turnMistral has released Mistral Large 4, its largest model to date: a 1-trillion-parameter, 52-billion-sparse-activation natively multimodal model trained from scratch rather than iterated from an earlier version. It is available in public preview on Mistral Studio now, with weights due at the end of the month.
- Heat index
- 25
- All sources
- 13
- All reports
- 20
- First seen
- 47 days ago
Reports on this story headlines open the original
Mistral has released Mistral Large 4, its largest model to date: a 1-trillion-parameter, 52-billion-sparse-activation natively multimodal model trained from scratch rather than iterated from an earlier version. It is available in public preview on Mistral Studio now, with weights due at the end of the month.
开源中国AI score 70A user on V2EX reports that Mistral Large 4 is now available, featuring a free Playground and OpenAI-compatible API for text and multimodal tasks. The model supports a 1M tokens context window, with pricing starting at $10 under a freemium model.
V2EXAI score 59
Artificial Analysis says Mistral has released Mistral Large 4, which scores 38 on the Artificial Analysis Intelligence Index and ranks as the most intelligent model from outside the US and China. Its results are described as comparable to DeepSeek V4.1 Flash (max).
TechmemeAI score 58SiliconANGLE reports that Mistral AI has opened access to Mistral Large 4, its most capable large language model to date, built on a 1-trillion-parameter mixture-of-experts architecture. It launches in public preview on the company's own cloud platform, with model weights promised later this month.
SiliconANGLEAI score 66Mistral has opened a public preview of Mistral Large 4, unofficially known as ML4 or Le Chonk. According to the company's blog, the model was trained on 3,800 Nvidia Grace Blackwell GPUs in its own data centers in Europe, and much of its training data was multilingual.
TechmemeAI score 61Mistral AI has released Mistral Large 4, nicknamed Le Chonk, in public preview: a 1.05T-parameter Mixture-of-Experts model with 49B active parameters, native image input and a 1M-token context window. It was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters; the API is live now and open weights are due at the end of October 2026.
MarkTechPostAI score 69
TechCrunch reports that French AI lab Mistral AI has released Mistral Large 4, a 1T-scale multimodal model positioned to leapfrog both American and Chinese rivals.
TechCrunch · AIAI score 67Mistral has released Large 4, its largest model yet, with one trillion parameters trained on its own European infrastructure. According to The Decoder, it makes a big leap on the independent Intelligence Index but still trails Claude, GPT-6 and Chinese rivals. Mistral is pitching the model for cybersecurity work that closed US models refuse to take on.
The DecoderAI score 71Mistral has launched a public preview of Mistral Large 4, a 1T-parameter natively multimodal mixture-of-experts model with 49B active parameters, nicknamed "Le Chonk." The API is available now through Mistral Studio, with open weights promised by the end of the month. Mistral reports 61.7% on DeepSWE v1.1, 93% on Cybench and 42% on Dense 200 visual grounding, narrowly above GPT-6 Astra's 41%.
TestingCatalogAI score 74Mistral AI has launched a public preview of Mistral Large 4 (ML4, codename le Chonk), available through the Mistral Studio API, with model weights to be released later this month. The natively multimodal model has 1 trillion total parameters and 49 billion active; Mistral says it reaches SOTA among open models on enterprise workloads such as cybersecurity, finance and legal, and red-teaming with security agencies and regulators is under way.
IT之家AI score 73Mistral released Mistral Large 4 (ML4, codename le Chonk) on Tuesday, a 1-trillion-parameter model the company says ranks among the world's top open models. cnBeta reports particular strength in cybersecurity, coding, manufacturing, finance and multimodal tasks.
cnBetaAI score 59Artificial Analysis says it has published the full benchmark results for Mistral Large 4 Preview. The post itself lists no scores and points readers to the accompanying analysis.
X:Artificial Analysis (@artificialanlys)First-partyAI score 40According to Artificial Analysis, Mistral Large 4 Preview scores 50 on its Cyber Index, level with GLM-5.3-Flash and behind MiMo-V2.6-Pro at 56. The firm says the model will rank among the top three open-weights models on the index once its weights are released.
X:Artificial Analysis (@artificialanlys)First-partyAI score 43Artificial Analysis reports that Mistral Large 4 Preview costs $1.13 per Intelligence Index task, based on standard pricing of $1.36 per 1M input tokens and $4.18 per 1M output tokens, with cached input at $0.14 per 1M. A 50% launch discount applies for the first two weeks.
X:Artificial Analysis (@artificialanlys)First-partyAI score 47According to Artificial Analysis, France now ranks third among countries for frontier model intelligence, behind the US (Claude Opus 5.5 at max effort, 58) and China (MiMo-V2.6-Pro, 46). The placement is based on the firm's Intelligence Index scores for each country's leading model.
X:Artificial Analysis (@artificialanlys)First-partyAI score 42Mistral has released Mistral Large 4 in Research Public Preview, scoring 38 on the Artificial Analysis Intelligence Index, with plans to release the weights later. Artificial Analysis says the release puts France back on top for the most intelligent model from outside the US and China.
X:Artificial Analysis (@artificialanlys)First-partyAI score 55Mistral AI has announced Mistral Large 4, a general-purpose multimodal model listed in its documentation at 1.05 trillion total parameters with 49 billion active, nicknamed Le Chonk. It is available in public preview through Mistral's API, with open weights planned for the end of October.
Unite.AIAI score 76Mistral has announced Mistral Large 4, codenamed Le Chonk, a 1T-parameter open-weight model with 49B active parameters and native multimodality. It rolls out through the API starting today, with the open-weight release planned for the end of October, and Mistral claims SOTA results on critical workloads including cyber.
X:TestingCatalog (@testingcatalog)AI score 70According to The Deep View, Mistral has released Mistral Large 4, codenamed 'le Chonk', a 1T-parameter open-weight model aimed at general agentic capabilities and trained on 4,000 Grace Blackwell GPUs. The report frames it as an attempt to break the dominance of closed flagship models from frontier labs.
TechmemeAI score 71Mistral has released a preview of Mistral Large 4, a 1-trillion-parameter model the Paris-based company calls the most powerful open-weight AI system outside China. Developers can access it through Mistral's API starting today, and the weights will be available for download and self-hosting in October.
The Next WebAI score 64
Other stories people are talking about
- Heat index 34Mistral opens Asia-Pacific HQ in Singapore, builds custom models1 sources
- Heat index 1134RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
- Heat index 1004Anthropic Mathematician Says Claude Formalized the Generalized Riemann Hypothesis21 sources
- Heat index 593Anthropic IPO total raise reaches $93.5 billion after greenshoe exercise19 sources
- Heat index 592RisingThinking Machines Lab Open-Sources Inkling 2 Mini with 32GB Model Card Recommendation10 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

