Mistral AI has opened access to Mistral Large 4, its most capable large language model to date, positioning it against frontier systems including GPT-6 Astra.
The model is now available in public preview through Mistral’s cloud platform, while the company plans to release its weights later this month.
1 trillion parameters, 49 billion active: Mistral Large 4
Mistral Large 4 uses a mixture-of-experts architecture with 1 trillion total parameters. It activates only 49 billion parameters at a time, making it more hardware-efficient than models that activate their full parameter count for every request.
Mistral says the model can answer questions across more than 160 languages. It was trained on 3,800 Nvidia Grace Blackwell chips.
Benchmarks: cyber and vision strengths
Mistral Large 4 earned a top-five score on the AA Cyber Index, which measures how well AI models can identify and fix software vulnerabilities. It performed particularly well at patching open-source software, scoring 82 per cent in that category and outperforming open-source rivals.
The company also highlighted computer vision as a strength. The model scored 1 per cent higher than GPT-6 Astra on Dense200, a benchmark that measures how well models identify objects of interest in images.
On coding, Large 4 scored higher than several leading open-source models, including Qwen 3.8 Max and DeepSeek V4 Pro, and beat DeepSeek V4 Pro on AutomationBench and AA-Briefcase. It still trails frontier models such as Astra on some of the industry’s most widely used coding benchmarks.
FAQs
What is Mistral Large 4?
Mistral AI’s most capable large language model, using a mixture-of-experts architecture with 1 trillion total parameters and 49 billion active at a time.
When will the model weights be released?
Mistral plans to release the weights later this month; the model is currently in public preview on its cloud platform.
What hardware trained it?
3,800 Nvidia Grace Blackwell chips, according to Mistral.
How does it compare with GPT-6 Astra?
It beats Astra by 1 per cent on the Dense200 vision benchmark but still trails frontier models on several widely used coding benchmarks.
It beats Astra by 1 per cent on the Dense200 vision benchmark but still trails frontier models on several widely used coding benchmarks.
The launch intensifies the race among AI labs, with each trying to outdo the other on reasoning benchmarks and enterprise adoption.
Related stories
- OpenAI Introduces World’s Smartest Model, ‘GPT-6 Astra’
- OpenAI turns ChatGPT into an interactive interface with Intelligent UI, alongside GPT-6
Originally reported by propakistani.pk.
