Mistral Large 4's weights are out — the 1T-parameter open model opens today

Mistral has released the weights for Mistral Large 4 as scheduled: a 1.05-trillion-parameter Mixture-of-Experts model with a 1M-token context, served from Mistral's own European data centres. Previewed on October 6 under the internal codename 'Le Chonk', it is the largest open-weights release of the year — and it lands hours after Anthropic's filing made the closed-vs-open split the industry's central question. Source: Mistral · Eden AI specs

Anthropic files its S-1: $16.3B in half-year revenue, an $80.7B net loss

Anthropic made it official on Monday, filing its Form S-1 to list on the Nasdaq under the ticker ANT. The filing shows $16.3 billion in revenue for the first half of 2026 — roughly thirteen and a half times the year-ago period — its first profitable quarter, and $265.6 billion in fixed compute commitments, against an $80.7 billion net loss that early analysts caution needs careful reading. The S-1 dominated Hacker News within hours. Source: Anthropic · Fortune · The VC Corner

SpaceX locks in $40B GPU financing — the largest ever — with Apollo

SpaceX has reached a binding commitment for an approximately $40 billion master finance lease for GPUs and data-centre equipment, disclosed in a regulatory filing on Monday. The structure matters: an affiliate of Apollo Global Management buys the chips and leases them to SpaceX, with deliveries expected in 2027 — the largest GPU financing in history is a sale-and-leaseback, putting a money manager between the chipmakers and their biggest new buyer. Source: Reuters · Apollo

🏠 Local models: no new qualifiers — but the DeepSeek quant train keeps rolling

Nothing new qualified for the local-models bar this cycle. The momentum story is last week's DeepSeek-V4.1-Flash GGUF quants nearly doubling again, from roughly 410k to about 890k downloads, with Nanbeige 4.3 33B also climbing to 600 likes — local inference demand keeps outpacing new releases. Source: Hugging Face · Nanbeige 4.3