Mistral AI launched Mistral Large 4 on Oct. 6, a 1-trillion-parameter mixture-of-experts model that activates 49 billion parameters per query. The Paris-based company said the model, nicknamed Le Chonk, costs $1.36 per million input tokens and $4.18 per million output tokens, significantly undercutting rivals like Claude Opus 5.5 and GPT-6 Astra.
The nickname follows the Le Chaton Fat meme from June. Mistral chief scientist Guillaume Lample wrote on X that Large 4 is “at the frontier of open models, and by far the strongest open-weight model from the US or Europe.”
On Surge AI’s blind human evaluation of coding quality, Large 4 ranked second with 3.74 out of 5, behind Claude Opus 5 at 4.22. On AutomationBench it scored 59.9, while Claude Sonnet 5.5 scored 71.8, Opus 5.5 scored 69.5 and Gemini 4 Argon topped the page at 77.5. On DeepSWE 1.1, Large 4 scored 62, ahead of GLM-5.3 and DeepSeek V4 Pro but behind Kimi K3’s 68.
Mistral says it will release Large 4’s weights by the end of October. The company positions itself as a provider of sovereign AI; Saudi Arabia’s state-backed HUMAIN signed a deal worth hundreds of millions of euros in August. In September, Mistral raised a €3 billion ($3.37 billion) Series D at a valuation above €21 billion ($23.6 billion), led by Samsung.
Separately, SpaceX is seeking about $40 billion with Apollo Global Management to acquire AI chips from Nvidia, according to the Financial Times. The financing would consist of approximately $10 billion in bank loans and $30 billion in investment-grade debt, with PIMCO among the lenders approached, and the transaction is expected to close in 2027. After the news, SpaceX stock dropped 1% in after-hours trading, while Nvidia gained 0.5%.
Nvidia said SpaceXAI will use Vera CPUs and the Vera Rubin platform as Grok’s infrastructure grows toward gigawatts of computing capacity. Vera has 88 Olympus cores and up to 1.2 TB/s of memory bandwidth. The first-generation Starmind satellite is set to use an optimized Vera Rubin NVL72 system. Each satellite has a peak processing capacity of 250 kilowatts, harnessing solar energy in sun-synchronous orbit.
Orbital AI competition is broadening: Google’s Project Suncatcher prototype carried four TPUs, while Starcloud has launched an Nvidia H100 into orbit. But the economics remain difficult. BCG estimates orbital data facilities cost 2.5 to 3 times more than ground facilities, and Brookings estimates one orbital data center would require 2.15 million square feet of radiators. Radiation-hardened chips are becoming more important for space computing.
Gartner projects worldwide AI spending will reach $2.7 trillion in 2026, up 49.5% year over year, while Morgan Stanley estimates AI infrastructure will need $1.5 trillion in outside financing by 2028.
