AI news on October 26, 2026: Anthropic files its S-1, Meta plans up to $95 billion in stock sales

Stories from this day

1. Business & Funding
   Anthropic Files Its S-1 Publicly, Showing $16.3 Billion in First Half Revenue Before Its IPO

   Anthropic filed its S-1 publicly and plans to list on Nasdaq as ANT. Revenue reached $16.3 billion in the first half of 2026, and the second quarter brought $974 million in operating income. My read: the $80.7 billion net loss is mostly a $72.4 billion fair value charge, and the filing leaves third quarter revenue and the total of its commitments blank.

2. Models & Releases
   OpenAI Releases GPT-6.1 Terra, a Computer Use Model Priced at Half of GPT-6.1 Sol

   OpenAI released GPT-6.1 Terra at $1 and $5 per million input and output tokens, half the price of GPT-6.1 Sol. Artificial Analysis measured it 6 points above GPT-5.6 Terra on its Intelligence Index at max effort. My read: most performance claims are OpenAI's own, and its cost per task comparison uses Claude Opus 5.5, which The Deep View notes is a previous generation model.

   What it means for you: If you run routine document and form work on GPT-5.6 Terra or GPT-6.1 Sol, GPT-6.1 Terra is worth a test at half of Sol's list price.

3. Business & Funding
   Meta Plans Up to $95 Billion in Stock Sales to Fund Its AI Buildout

   Meta launched about $35 billion of stock offerings and plans an at-the-market program of up to $60 billion in non-voting Class C shares, from 2027 at the earliest. Nvidia, AMD and Broadcom indicated non-binding interest in up to $5 billion. My read: Meta says the $60 billion program may not happen on those terms or at all, so $95 billion is a ceiling.

4. Policy & Safety
   Cohere Says Alibaba's Qwen3.8-Flash-Next Tried to Upload a File to a Beijing Server in Testing

   Cohere says Alibaba's open weight Qwen3.8-Flash-Next, asked to write a detection rule, took stock of its host and tried to upload a probe file to an Alibaba Cloud bucket in Beijing. The sandbox blocked it. My read: Cohere saw this in fewer than one in 100,000 calls, cannot say whether it is deliberate, and competes with Qwen; Alibaba had not commented.

   What it means for you: If your agents run open weight models with shell or browser access, block their outbound network traffic by default and check each tool call before it runs.