News
AI news today: Anthropic files for IPO, OpenAI ships GPT-6.1 Terra
Anthropic's S-1, a cheaper OpenAI computer-use model, Meta and SpaceX financing for AI compute, and a new Transformers release.
Tuesday’s roundup: Monday was all about money and models. Anthropic opened its books in a public IPO filing, OpenAI cut the price of computer-use AI with GPT-6.1 Terra, and Meta and SpaceX each lined up huge financing for AI compute. Here’s what happened and why it matters.
Anthropic files to go public as “ANT”
In its S-1 filing, Anthropic reported $16.3 billion in first-half 2026 revenue and an $80.7 billion net loss. Most of that loss, $72.4 billion, came from a change in the fair value of convertible notes, derivative liabilities and SAFEs, which the S-1 says was driven by Anthropic’s rising valuation. TechCrunch calls these noncash accounting adjustments (TechCrunch). The share count and price range haven’t been set yet.
Why it matters: It’s the first full public look at a frontier AI lab’s finances, and it sets the bar for every other lab that wants to go public.
OpenAI launches GPT-6.1 Terra
It’s a fast model built for computer use, and OpenAI says API prices are 50% lower than GPT-5.6 Terra. The benchmark claims come from OpenAI itself (OpenAI).
Why it matters: Cheaper computer-use models make high-volume automation, like entering invoices into accounting software, more practical for developers.
Meta plans to raise up to $95 billion
It starts with about $35 billion in stock offerings now, plus a planned $60 billion at-the-market program whose terms aren’t set yet. The money will fund AI infrastructure and what Meta calls “personal superintelligence”. Nvidia, AMD and Broadcom have indicated non-binding interest in buying up to $5 billion of the shares, according to Meta’s SEC filing and CNBC. Details are in Meta’s announcement.
Why it matters: Even one of tech’s biggest cash generators is now selling stock to pay for AI infrastructure.
SpaceX commits to a roughly $40 billion GPU lease with Apollo
The deal covers Nvidia Vera Rubin hardware, with leases expected to start in the first half of 2027 (SpaceX).
Why it matters: SpaceX calls it the largest committed GPU financing, and it shows AI chips being leased rather than bought outright.
Hugging Face Transformers v5.21 is out
It adds models like MiniCPM-OCR, Flint and Pocket TTS, plus an FP8 KV cache. Some tensor-parallel and MoE changes will break existing setups (GitHub).
Why it matters: New open OCR, speech and MoE models are now in the standard toolkit, but teams with custom parallel plans should check the breaking changes before upgrading.