Mistral AI Drops Mistral Large 4 Model Le Chonk

Mistral AI Unveils Mistral Large 4
Paris-based AI innovator Mistral AI officially launched its newest flagship model, Mistral Large 4, on Tuesday, October 6. The system boasts a massive 1 trillion parameters.
Parameters are essentially the adjustable dials that an AI tunes during its training. Having more parameters generally boosts a model's ability to learn complex ideas, though it makes running the system more expensive.
Despite its serious engineering, Mistral Large 4 carries a humorous official nickname: Le Chonk. The playful moniker grew out of a viral internet meme that won over the developer community.
How Mixture of Experts Keeps Costs Down
Running 1 trillion parameters for every simple question would be wildly expensive. Mistral solves this challenge by using a design known as a mixture of experts.
Think of this structure like a well-organized company with specialized departments. When you ask a question, the system routes your query only to the specific specialist networks needed for that task.
Thanks to this setup, only 49 billion parameters fire per query. That marks a steady increase from Mistral Large 3, which activated 41 billion parameters out of 675 billion total parameters.
The Internet Meme Behind the Name Le Chonk
The backstory behind 'Le Chonk' shows how tech companies interact with community humor. Back in June, Mistral renamed its Le Chat assistant to Vibe.
In response, community members on Reddit and X playfully invented a fictional AI model called Le Chaton Fat—translated roughly as 'the fat kitten'.
Fans created joke benchmark posts claiming this imaginary model had 30 trillion parameters and outputted 1,000 meows per second, outperforming competitor systems like Claude Fable 5.
Mistral AI Chief Executive Officer Arthur Mensch joined in on the joke, replying that the system was actually named 'le gros chaton' (French for 'the big kitten').
The team even placed a cartoon cat on its Vibe product website. With this release, Mistral's official announcement post formally declared the model's nickname as 'le Chonk'.
High Performance at a Fraction of the Cost
AI companies bill developers using tokens. A token is a tiny chunk of text, equal to roughly three-quarters of a word.
Mistral Large 4 comes with highly competitive pricing. It costs $1.36 per million input tokens and $4.18 per million output tokens.
Here is how that price tag compares directly to leading industry rivals:
- Mistral Large 4: $1.36 input / $4.18 output per million tokens
- Claude Opus 5.5: $4.00 input / $20.00 output per million tokens
- GPT-6 Astra: $10.00 input / $50.00 output per million tokens
That makes Large 4 roughly one-third the input cost and one-fifth the output cost of Claude Opus 5.5. Against GPT-6 Astra, it comes in at roughly one-seventh the input price and one-twelfth the output price.
Sovereign AI and Huge Capital Investment
Mistral focuses heavily on building sovereign AI. These are models that nations or individual enterprises can host on their own infrastructure without sending private data to outside corporations.
This privacy-focused strategy has generated major business deals. In August, Saudi Arabia's state-backed firm HUMAIN signed a agreement worth hundreds of millions of euros with Mistral.
Additionally, Mistral closed a €3 billion ($3.37 billion) Series D funding round in September led by Samsung, valuing the developer above €21 billion ($23.6 billion).
Mistral stated that Large 4 represents the first major product milestone funded by that massive capital raise.
Open-Weight Release and Chief Scientist Statements
Mistral plans to release the model's open weights by the end of October. Open-weight means programmers worldwide can download the trained files to run tests locally.
"[Mistral Large 4] is at the frontier of open models, and by far the strongest open-weight model from the US or Europe," wrote Mistral AI Chief Scientist Guillaume Lample on X.
Releasing open weights gives independent analysts and research teams the freedom to test all claimed benchmarks themselves.
Detailed Benchmark Performance Breakdown
In its launch announcement, Mistral primarily benchmarked Large 4 against popular Chinese open-weight models such as DeepSeek V4 Pro, Kimi K3, GLM-5.3, and Qwen3.8 Max.
Closed models like Claude and GPT appeared in select performance comparisons across various standardized tests:
Finance Agent v2 Scorecard
On the Finance Agent v2 benchmark, Mistral Large 4 recorded a score of 54.7. This placed it ahead of GPT-6 Astra at 53.5, but behind Claude Opus 5.5 at 58.6.
Coding Benchmarks: Surge AI and Deep SWE 1.1
During a blind human evaluation of coding quality conducted by Surge AI, Large 4 ranked second out of five models with a score of 3.74 out of 5, trailing Claude Opus 5 at 4.22.
On the Deep SWE 1.1 software development test, Large 4 posted a score of 62. That tops GLM-5.3 at 61 and DeepSeek V4 Pro at 57, but trails Kimi K3 at 68.
By comparison, Datacurve's overall coding leaderboard places GPT-6 Astra and Claude Opus 5 tied at 74.
AutomationBench Real-World Tasks
AutomationBench evaluates AI systems by giving them 657 simulated office chores across finance, HR, sales, and support. Models get zero points if they violate operational guidelines.
Mistral Large 4 scored 59.9 points on AutomationBench. On Artificial Analysis's tracking board, Claude Sonnet 5.5 hit 71.8, Opus 5.5 reached 69.5, and Gemini 4 Argon led at 77.5.
Crypto Markets and Tech Culture Overview
The world of open-source AI models, online meme culture, and digital assets often overlap in community engagement.
Looking at broader market activity, BitcoinBTC-0.35% was trading at $85,614 (+3.05%), with a 24-hour high of $86,648, a low of $85,122, and $1.1 billion in daily volume.
Past performance of digital assets or market indices is not an indicator of future results. Always evaluate risk factors before participating in any market opportunity.
Latest blog posts

Mark Moss Sees Strong Bitcoin Outlook Driven by ETFs
Bitcoin Opportunity Fund co-founder Mark Moss suggests that expanded access via banks and ETFs could set the stage for a new record high by 2027.

Poland Lower House Backs Crypto Sanctions Bill
Poland's lower house has approved a bill expanding national sanctions lists and authorizing crypto asset freezes, moving the proposal to the Senate.

Vitalik Buterin Recasts ETH as a Store of Value
Vitalik Buterin reframes Ethereum as a store of value alongside prediction market insights on ETH price thresholds.