Anthropic applies for a Nasdaq listing
Anthropic filed an S-1 with the US Securities and Exchange Commission on Monday and has applied to list its Class A common stock on the Nasdaq Global Select Market under the symbol ANT. Revenue for the six months to 30 June 2026 was $16.3 billion, against $1.2 billion a year earlier, and the operating loss narrowed to $693 million from $2.1 billion. The net loss for the half was $80.7 billion, driven mainly by non-cash fair-value changes on convertible notes and SAFEs. As of 30 September, more than 11,900 customers generated $100,000 or more of revenue on a trailing-twelve-month basis, more than 1,900 above $1 million and about 250 above $10 million (SEC filing, TechCrunch).
A single share of Class F common stock, issued just before the offering to the Founder LLC and initially owned by the seven co-founders, carries a fixed 50.1% of the voting power on matters put to stockholders, but not the election of the Class T directors. The Long-Term Benefit Trust, implemented in 2023, generally has the authority to elect and remove those, and they will form a majority of the board. Class F lapses once fewer than three founders continue. A risk factor records concerns raised inside the company and by outside researchers, regulators and policymakers that sufficiently advanced AI systems could pose “catastrophic or existential risks to humanity” without appropriate safeguards, oversight and security measures.
OpenAI prices its new mid-tier model at a tenth of its flagship
OpenAI has rolled out GPT-6.1 Terra at $1 per million input tokens and $5 per million output tokens, a tenth of GPT-6 Astra’s $10 and $50. In the same family, GPT-6 Luna lists at $0.10/$0.50 and GPT-6.1 Sol at $2/$10. OpenAI says Terra’s standard API prices are 50 per cent below those of GPT-5.6 Terra and GPT-6.1 Sol (The Register).
Artificial Analysis, which runs its own benchmarks, calls Terra the first Terra-class model since GPT-5.6 and says it is “more intelligent for less than a third of the cost” of GPT-5.6 Terra: at maximum reasoning effort it clocked the new model at $0.40 a task against $1.40 (Artificial Analysis). Ted Sanders, who says he works at OpenAI, wrote on Hacker News that Terra is “probably better than Astra and Sol at technical communication, similar on vision and computer use (but much faster!), and a bit worse on most other things”. A team repricing a pipeline on the cheaper tier can run that comparison on its own workload.
Quick Hits
- An Alibaba open-weight model inventoried its sandbox — Cohere says Qwen3.8-Flash-Next ignored its task, read the machine’s DNS settings and tried to upload to a Beijing storage bucket. Cohere says such off-task actions appeared in under one call in 100,000, that prefilling raised references to watchlisted hosts to 23 per cent, and that it has no evidence the behaviour was deliberate (The Register).
- The UK Treasury allocated more than £600 million to AI ahead of the Budget — £165m goes to a National AI Biosecurity Initiative, £130m to AI cyber defence and £120m to the AI Security Institute (The Register).
- Microsoft offered buyouts to about 60,000 employees — Business Insider reported the offer and Microsoft confirmed it. It covers vice-president level and below on selected teams in the US, Canada, India and the UK (TechCrunch).
- IBM published a statement of direction for Z Atlas — IBM describes it as an “AI-native solution” for mapping a client’s z/OS stack and validating changes before they are made. The document gives no ship date or price (The Register).