Mistral Large 4: Inside the 1T-Parameter Flagship Launch

Episode 1 · Mistral Unveils Mistral Large 4, a 1T-Parameter Open Multimodal Model (2026-10-06, 42 posts)

Mistral officially announced its flagship open-weight model Mistral Large 4 (codename Le Chonk) on October 6: a MoE architecture with 1T total parameters and roughly 49B active parameters, natively multimodal. The API launched the same day, with open weights planned for late October.

Confirmed

• Specs: 1T total parameters, 49B active parameters, natively multimodal; its visual grounding capability is claimed to surpass closed-source models.
• Positioning: billed as the strongest open-weight model in the US and Europe on aggregate benchmarks, achieving SOTA on key workloads such as cybersecurity defense, manufacturing, and finance.
• Rollout: the API went live on launch day; open weights are slated for release in late October.

Unconfirmed

• @haider1 relayed community skepticism, noting gaps in the officially released benchmark data—actual performance awaits independent evaluation.

Why it matters

• This is Mistral's largest open-weight flagship to date, and a 1T-parameter model is a landmark for the European open-source camp, seen as a move to rival top US open models.
• If its SOTA claims in cybersecurity and other critical industries hold up under third-party validation, it could reshape enterprise model selection.

• Mistral launches 1T-parameter MoE model Large 4, claims best open-weights model in US/Europe — Bam4d · 2026-10-06
• Mistral launches Le Chonk: 1T-parameter multimodal model, open weights due end of October — MistralAI · 2026-10-06
• Mistral Ships Large 4, and the Community Has Already Nicknamed It 'Le Chonk' — cpldcpu · 2026-10-06
• Mistral unveils Mistral Large 4 Le Chonk: a 1T-parameter open-weight model with 49B active — testingcatalog · 2026-10-06
• Mistral launches 1T-parameter open-weights flagship Mistral Large 4 — reach_vb · 2026-10-06
• Mistral Large 4 Preview Benchmarks Just Dropped — Informal-Trouble2183 · 2026-10-06
• Mistral launches 1T-param Mistral Large 4; users question its benchmark claims — haider1 · 2026-10-06
• Mistral's ML4 Hits Open-Weight SOTA on Security, Scores 82% on Vulnerability Patching — GuillaumeLample · 2026-10-06
• Mistral ML4 Benchmarks: SOTA on Finance and Legal Workflows, Terminal and Spreadsheet Navigation — GuillaumeLample · 2026-10-06
• Mistral ML4 Human Eval Beats GLM 5.3 on STEM, CAD, Finance; Ties on Agentic Coding — GuillaumeLample · 2026-10-06
• Mistral ships 1T-parameter open-weight ML4, beating GLM 5.3 on human eval, hiring worldwide — GuillaumeLample · 2026-10-06
• Mistral releases 1T-parameter 'Le Chonk', claims best open-weight model outside China — nordicinst · 2026-10-06
• Mistral launches 1T-parameter multimodal Large 4 preview, open weights by month-end — qtnx_ · 2026-10-06
• Mistral launches Large 4 preview: 1T-param multimodal model with 49B active, open weights due October — arthurmensch · 2026-10-06
• Mistral Launches Large 4: 1T-Parameter Multimodal MoE With 49B Active Params, Open Weights in October — MaybeLiterally · 2026-10-06
• Mistral Large 4 leaked on Hugging Face: 1T total / 52B active params, open-source this month — victormustar · 2026-10-06
• Mistral's Sophia Yang: new model beats GLM 5.3 on human evals across domains — sophiamyang · 2026-10-06
• Mistral launches Mistral Large 4 (Preview) — Few_Painter_5588 · 2026-10-06
• Mistral to open-source LeChonk-1T-A49B, weights due end of month — wapswaps · 2026-10-06
• Mistral officially launches Mistral Large 4 in Research Public Preview — MistralAI · 2026-10-06

22 more related posts →

Episode 2 · Mistral Trains 1T-Parameter Model on 3,800 Grace Blackwell Chips in Europe (2026-10-06, 3 posts)

Mistral trained its 1-trillion-parameter Mistral Large 4 end-to-end (pretraining to post-training) on 3,800 NVIDIA Grace Blackwell superchips at its European data center, funded by a €3B Series D, with two more clusters on the way.

• ML4 Trained on 3,800 Grace Blackwell GPUs; Mistral's C and D Round Clusters Coming Online — GuillaumeLample · 2026-10-06
• 1T-parameter model fully pre- and post-trained on 3,800 Grace Blackwells in Europe — qtnx_ · 2026-10-06
• Mistral Large 4 trained on nearly 4,000 NVIDIA Grace Blackwell Superchips, backed by €3B Series D — cedric_chee · 2026-10-06

Episode 3 · Mistral Large 4 Still in RL Training, Expected by Month's End (2026-10-06, 3 posts)

Mistral Large 4 (codenamed "le chonk") is still undergoing RL training with continuous improvements over its preview, and is expected to be released for wider use later this month, according to leaks and official channels.

• Mistral Large 4 Preview Still Improving as RL Run Continues, Launch Set for This Month — qtnx_ · 2026-10-06
• Mistral Large 4 preview: RL run still improving, launch later this month — xeophon · 2026-10-06
• Mistral Large 4 still in RL training, seeing gains, release due end of month — DerpSenpai · 2026-10-06

Episode 4 · Artificial Analysis Benchmarks Mistral Large 4: France Rises to Third on Frontier Leaderboard (2026-10-06, 6 posts)

On October 6, Artificial Analysis published a full review of the Mistral Large 4 Research Public Preview: the model has 1T total parameters with 49B active, scores 38 on the Intelligence Index—placing it in the same tier as and close to Gemini/GPT—and its weights are slated to be open-sourced by the end of the month. The review calls it a landmark marking France's return to the strongest model camp outside the US and China, with open-source dynamics and pricing strategy worth watching.

Confirmed

• Model specs: 1T parameters, 49B active, 38 on the Intelligence Index, in line with GP… (remaining metrics in the full review match the main post)
• Country rankings: with 38 points, France rises to third on the frontier-model nation leaderboard, behind only the US (Claude Opus 5.5, 58 points) and China (MiMo-V2.6-Pro, 46 points)
• Pricing: $1.36/$4.18 per million input/output tokens, with cached input at $0.14; 50% off for the first two weeks after launch, bringing per-task cost to $0.57—but full-task cost is about $1.13, still 4x that of comparable open-source models and higher than competitors like GLM-5.3
• Cybersecurity capability: 50 on the Cyber Index, tied with GLM-5.3-Flash and below MiMo-V2.6-Pro (56); the reviewers expect its Cyber ranking to enter the top three among open-source models once the weights ship
• Artificial Analysis has launched a full review results page covering all capability metrics

Why It Matters

• If the 1T-scale weights are open-sourced on schedule at month's end, it will significantly reshape the open-source model landscape and establish France as a third pole beyond the US and China
• The high pricing has sparked value-for-money debate: leading capability but several times the cost of peer open-source models may limit real-world adoption
• It still trails China's top open-source models (MiMo-V2.6-Pro) in specialized capabilities like cybersecurity, and post-release ranking shifts are worth tracking

• Mistral Large 4 scores 38 on AA Intelligence Index, 1T open weights due end of October — ArtificialAnlys · 2026-10-06
• France ranks third by country for frontier model intelligence on AA index — ArtificialAnlys · 2026-10-06
• Mistral Large 4 pricing: $1.13 per task, 4x costier than similar open weights models — ArtificialAnlys · 2026-10-06
• Mistral Large 4 scores 50 on Cyber Index, set for top-three open weights spot — ArtificialAnlys · 2026-10-06
• Full benchmark results for Mistral Large 4 Preview published by Artificial Analysis — ArtificialAnlys · 2026-10-06
• Mistral Large 4 scores just 38 on AA intelligence index, panned for a 1T model — realmrfakename · 2026-10-06

Episode 5 · Mistral Large 4.0 Surfaces on Hugging Face, Weights to Open-Source This Month (2026-10-06, 3 posts)

A Hugging Face preview page reveals Mistral Large 4.0 (codenamed Le Chonk), a natively multimodal model with 1 trillion total parameters and about 52B active. Weights are expected to be open-sourced around the end of this month.

• Mistral-Large-4.0 weights (1T total, 52B active) expected in ~25 days — paf1138 · 2026-10-06
• Mistral Large 4 'Le Chonk' teased on Hugging Face: 1T params, 52B active, open weights Oct 31 — LysandreJik · 2026-10-06
• Mistral Large 4 'Le Chonk' leak: 1T-parameter multimodal model with open weights Oct 31 — onetwoval · 2026-10-06