AI news, October 27, 2026: GPT-6.1 Terra, Anthropic files for its IPO, Meta's $95B share sale
Five stories from Monday, October 26, 2026 (US time) for solo makers who build with AI models and tools. Two are about what you can call from your code today; the other three show how much money is now flowing into AI compute. Figures are as the companies and outlets state them.
OpenAI releases GPT-6.1 Terra, a cheaper model for computer use and everyday coding
OpenAI introduced GPT-6.1 Terra, a fast mid-range model aimed at visual reasoning, operating desktop apps, writing and coding. The API model gpt-6.1-terra is listed at $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% below GPT-5.6 Terra and GPT-6.1 Sol. In OpenAI's own tests it scores 50.3% on Terminal-Bench 4.0 at maximum effort, up from 21.5% for GPT-5.6 Terra, though the company still recommends its larger GPT-6 models for the hardest coding work. It showed up the same day on Amazon Bedrock, Vercel AI Gateway and GitHub Copilot.
Why it matters for makers: high-volume jobs like reading invoices or filling forms just got cheaper to automate, but the benchmark numbers are OpenAI's own until independent tests arrive.
Sources: OpenAI, GitHub Changelog, Vercel changelog, AWS
Anthropic files its S-1 to list on Nasdaq as "ANT"
After weeks of leaked drafts, Anthropic filed for an initial public offering and plans to trade on Nasdaq under the ticker ANT. The filing shows $16.3 billion in revenue for the first half of 2026, up from $1.2 billion a year earlier, and an operating loss of $693 million; the $80.7 billion net loss is largely a non-cash charge tied to convertible debt, TechCrunch reports. Anthropic says more than half of customers spending at least $100,000 a year now use both its API platform and an app such as Claude Code, and its seven co-founders will keep 50.1% of the voting power.
Why it matters for makers: the filing lists cloud partners turning into competitors and dependence on its largest customers as risks, useful context if your product sits on a single model vendor.
Sources: TechCrunch, TechCrunch on enterprise strategy, CNBC
Meta plans to raise $95 billion in new stock to pay for AI
Meta said it will raise $35 billion through underwritten public share offerings and up to another $60 billion later through an at-the-market program. Nvidia, AMD and Broadcom have shown interest in buying up to $5 billion of the shares between them. Meta expects capital spending of $135 billion to $145 billion this year and says the money supports its apps, its Muse AI agent and its data center business.
New lab Mirendil debuts Fern, a model with a 10M-token context window
Mirendil announced Fern, its first model, pretrained from scratch with support for up to 10 million tokens of context. The company says that is roughly 23 MB of code, or about half a million lines, held at once. The first version, Fern 0.1, had only a short round of post-training and is pitched at reading whole repos, long training logs and full agent traces, with more long-context results promised later.
Why it matters for makers: very long context could make "read the entire codebase and its logs" a single prompt, but these are launch claims from a brand-new lab, so wait for outside testing.
Source: Mirendil
SpaceX signs a $40 billion GPU lease deal with Apollo
SpaceX disclosed a binding commitment under which an Apollo-managed entity will buy roughly $40 billion of AI chips and server hardware and lease it to SpaceX for about five years. Apollo said the deal covers Nvidia's next-generation Vera Rubin chips, and leases are expected to start in the first half of 2027. TechCrunch notes SpaceX's AI segment, which includes Grok and Cursor, brought in $7.6 billion of its $14.6 billion third-quarter revenue.
Source: TechCrunch