Skip to content
NewA calmer, faster ZeroHour
StoryModel releaseinfo

Mistral releases open weights for trillion-parameter Large 4

5 sources · 5 articlesFirst reported Updated ()1
What's new

October 30: CSO Online reports open weights fall outside API guardrail updates, Mistral has not addressed the EU 10^25 FLOP threshold, and a partner pen-test variant is in preview.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Mistral has published Mistral Large 4 weights, while open-weight cyber guardrails remain limited to API and hosted users.

French company Mistral published open weights for Mistral Large 4, nicknamed Le Chonk, on Hugging Face and through its API after a public preview earlier in October. The company describes a roughly 1-trillion-parameter natively multimodal model, about 52 billion parameters active per step, trained from scratch on European infrastructure, with low, medium, and high reasoning modes. Mistral says further reinforcement learning improved coding, defensive cybersecurity, and token efficiency, reporting a 51% Artificial Analysis Cyber Index score and a 96.6% refusal rate on malicious cyber requests. The Decoder says high mode scores 26 on Intelligence Index v5, below DeepSeek V4.1 Flash and GLM-5.3 at 32. CSO Online reports that downloaded weights sit outside updated API guardrails and that Mistral has not said whether the model meets the EU systemic-risk FLOP threshold.

Key facts
  • About 1 trillion parameters, with 52 billion active per step.
  • API pricing is $0.70 per million input tokens and $2.10 output.
  • Modified MIT license; model-as-a-service needs a separate license.
  • Mistral reports 96.6% refusal of malicious cyber prompts.
  • A vetted penetration-testing variant is in partner preview.
  • SGLang docs say a standard build needs eight Nvidia B300 GPUs.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

Timeline

dated events in the reporting · oldest first
  1. Oct 29, 2026Mistral publishes Large 4 weights on Hugging Face and its API. · TechCrunch
  2. Oct 30, 2026CSO reports open weights remain outside updated API guardrails. · CSO Online

What each outlet adds

beyond the shared facts
  • TechCrunchMistral claims 47 percent fewer tokens than GLM-5.3.
  • Hacker NewsReports three reasoning modes and a 48% coding-agent score.
  • The DecoderIntelligence Index 26, eight-GPU requirement, and token-price details.
  • CSO OnlineWeights miss API guardrails; EU FLOP status unstated.
Vendors
MistralMistral AI
Organizations
OpenAIAnthropicHugging FaceArtificial AnalysisOrange CyberdefenseENISAUK AI Security Institute
Countries
FranceChina
AI models
Mistral Large 4GLM-5.3Le ChonkClaude Opus 5.5DeepSeek V4.1 FlashMiMo-V2.6-ProKimi K3.1 Pro

Coverage

oldest first · one row per article · UTC
  1. · 1d ago Hugging Face BlogFirst report

    Mistral Large 4, nicknamed Le Chonk, on Hugging Face

    The post announces the model on Hugging Face; no license, weights, or benchmark details are in the captured item.

  2. · 1d ago TechCrunch · AI

    Mistral releases open weights for trillion-parameter Mistral Large 4

    Nicknamed Le Chonk, ML4 is on Hugging Face and Mistral's API, and the company claims 47 percent fewer tokens than GLM-5.3.

  3. · 1d ago Hacker News · AI

    Mistral releases open weights for Mistral Large 4

    The 1-trillion-parameter multimodal model adds three reasoning modes and reported defensive-security gains.

  4. · 1d ago The Decoder

    Mistral releases Large 4 open weights, nicknamed Le Chonk

    The trillion-parameter model leads some open coding scores but trails top Chinese models and Claude overall.

  5. · 10h ago CSO Online

    Mistral open-sources Large 4 weights despite cyberattack uplift warning

    Downloaded weights sit outside API guardrail updates, and Mistral has not said whether Large 4 meets the EU systemic-risk threshold.