France’s Mistral AI has released the weights of Mistral Large 4, its flagship one-trillion-parameter model nicknamed “Le Chonk”, making it one of the largest and most capable AI models that anyone can download, inspect, fine-tune and run on their own servers. The release, which landed on Hugging Face on 29 October, comes a few weeks after the model entered public preview earlier in October, and Mistral says three weeks of additional reinforcement learning since then have pushed it to the front of the open-weight pack on coding and defensive cybersecurity.

For India, where policymakers and startups are debating how much of the country’s AI stack must be built at home, the launch is a timely case study in what “sovereign AI” looks like when a mid-sized national champion takes on the American and Chinese giants.

What Mistral has released

According to the model card, Mistral Large 4 is a natively multimodal Mixture-of-Experts (MoE) model with about one trillion total parameters, of which roughly 49 billion are active for each token (about 52 billion counting embedding and output layers). It uses 128 routed experts, with four active per token plus one shared expert, accepts both text and images, and supports a context window of one million tokens. Mistral says it was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own data centres in Europe.

The model combines instruction-following, step-by-step reasoning, coding and agentic tool use in a single set of weights, and now offers three reasoning settings, low, medium and high, so developers can trade depth of reasoning against speed and cost.

The benchmark claims

Mistral has published a long list of results, all company-reported:

• Coding agents: 48% on the Artificial Analysis Coding Agent Index, effectively level with China’s GLM-5.3, the top open-weight model on that index, while using 47% fewer tokens per task on average.
• Terminal work: 41% on Terminal-Bench 4.0, up 13 percentage points from the preview, with 30% fewer output tokens.
• Real-world code quality: 19.7% on Cognition’s private FrontierCode 2.0 benchmark, which judges whether a repository maintainer would actually merge the agent’s change.
• Cybersecurity: 51% on the Artificial Analysis Cyber Index, alongside a 96.6% average refusal rate on malicious cyber requests.
• Enterprise tasks: 73.3% on the Finch finance-workflow benchmark and 84.5% on BrowseComp for web research.

Mistral also acknowledges a limit. On the broader Artificial Analysis Intelligence Index, Large 4 scores 26, the highest among Western open-weight models, but the leading Chinese open models still score higher on that aggregate measure.

Open weights, with one commercial catch

The model ships under what Mistral calls the Mistral Ecosystem Open-Weight licence, a modified MIT licence. Organisations can use, modify, fine-tune and commercialise it freely, with one condition: any business that sells the model itself as a hosted service must take a separate commercial licence from Mistral. For companies that prefer not to run it themselves, Mistral’s API prices it at $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens. It is also available through Microsoft Foundry, NVIDIA-hosted endpoints and OpenRouter, and can be self-hosted with vLLM or SGLang.

Mistral has been unusually candid about safety. It says that while monitoring live traffic during the preview, it saw “a potential for uplift to conduct real attacks” and is tightening its guardrails. A more capable cybersecurity variant is available only to vetted enterprise partners, under contract and for testing on systems they own or are authorised to probe.

Why India should pay attention

Mistral pitches Large 4 squarely as a sovereignty play: a model with “clear lineage from inception to release” that governments, banks and regulated industries can host on their own infrastructure without routing data through a foreign API. That argument resonates in New Delhi. Prime Minister Narendra Modi met Mistral co-founder and CEO Arthur Mensch in Paris in June, and India’s own push, from the IndiaAI Mission’s subsidised GPU pool to Sarvam AI’s plans for a trillion-parameter model, rests on the same idea that critical AI should not be a black box rented from abroad.

The release offers Indian builders three practical lessons.

• Scale is reachable, but not cheap. Mistral needed thousands of top-end Blackwell GPUs and its own data centres to train a trillion-parameter model. India’s shared national compute is growing fast, but home-grown frontier training at this scale still requires dedicated, concentrated clusters.
• Open weights change the build-or-adapt maths. Indian enterprises and startups can now take a frontier-class model, host it inside India and fine-tune it on their own data for legal, finance or citizen-service use cases. The model card lists Hindi, Bengali and Nepali among its supported languages, though real-world quality across India’s many languages and code-mixed speech still needs local testing.
• Sovereignty is about the whole stack. Downloading weights does not, on its own, deliver independence. Hardware supply, data residency, security of the software supply chain and the skills to run and improve these systems matter as much as the model.

The Elets CIO view

For Indian CIOs, Mistral Large 4 widens the menu. A credible, openly licensed alternative to closed American models and fast-improving Chinese ones means more negotiating power, more deployment options inside Indian data centres and an easier path to meeting data-localisation rules. The harder question is strategic: whether India’s own model builders can match this pace, or whether the country’s sovereign AI future will be built on adapting the best open models from elsewhere. Either way, the bar for “sovereign” just moved.

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!