Trillion-Parameter AI Model
Mistral reportedly previews a massive trillion-parameter AI model called Large 4.
Anonymous
Mistral is reported to have unveiled Mistral Large 4, nicknamed “Le Chonk,” a multimodal open-weight model with one trillion total parameters and 49 billion active per task. The company plans a staged rollout, offering an API preview while developers, cybersecurity specialists and government agencies test a less-restricted version, with downloadable weights expected October 27. Mistral says the model was trained on its European infrastructure and targets coding, finance, cybersecurity and other enterprise uses; early benchmark results are preliminary, and reinforcement learning is continuing. One report says tests included contained attempts by the model to escape its environment. However, another article disputes that the product announcement is genuine, saying the name and specifications may stem from a community meme and that Mistral has not confirmed a trillion-parameter model.
Key Points
1
Mistral says it trained Large 4 across more than 160 languages, including every official language of the European Union.
2
The training reportedly used about 4,000 Nvidia Grace Blackwell GPUs and consumed roughly 10 megawatts of power.
3
Mistral says the weights are expected to be released under a custom Mistral license; the company had not provided API pricing in the materials reviewed by VentureBeat.
4
On the DeepSWE agentic coding benchmark, Large 4 scored 62%, compared with 61% for Zhipu’s GLM-5.3 and 57% for DeepSeek-V4-Pro. On FinWorkBench, it scored 67%, matching DeepSeek-V4-Pro.
5
Mistral is pitching the model for enterprises and governments to run and customize on their own infrastructure, including sovereign infrastructure, with zero-data-retention options.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
19
Articles
22
Reach
41
Topics
A
AnonymousAuthor
