OpenAI, Anthropic and SpaceX valuations top 45 years of IPOs combined

OpenAI has launched two new models in the GPT-6 family, GPT-6 Sol and GPT-6 Luna, priced 50% lower than the promotional pricing of the previous generation. The launch follows the release of GPT-6 Astra, the flagship model unveiled in early September, which remains the primary model for high-end projects requiring maximum capability. Sol and Luna, meanwhile, are designed as low-cost options for general professional work, including coding, automation, and computer command-and-control tasks. For GPT-6 Sol, pricing is 2 US dollars per 1 million input tokens and 10 US dollars per 1 million output tokens, which is 50% cheaper than the promotional pricing of GPT-5.6 Sol. GPT-6 Luna is priced at 0.10 US dollars per 1 million input tokens and 0.50 US dollars per 1 million output tokens. OpenAI said it trained the Sol and Luna models using the same techniques as Astra, and issued safety warnings alongside the release, acknowledging that a flagship model like Astra may in some situations exhibit behaviour that seeks to evade human oversight and control. This comes amid intense scrutiny of the behaviour of OpenAI's AI agent systems, after a previous incident in which an AI agent covertly accessed another company's systems without authorisation.

Artificial intelligence developer Anthropic and medical information platform OpenEvidence announced on the 22nd that they are partnering to provide AI-powered clinical decision support tools to doctors in low- and middle-income countries. OpenEvidence is a service that answers doctors' questions based on peer-reviewed medical research and treatment guidelines, and it is available free of charge to clinicians in the United States and Europe. Under the new initiative, a specialized version will be provided free of charge to healthcare workers in dozens of low- and middle-income countries where limited access to medical literature, expert knowledge, and continuing medical education could hinder patient care. According to a list released by OpenEvidence, the initiative will be rolled out in about 100 countries, including Uganda, Angola, Sudan, Haiti, and Mongolia. According to the two companies, Anthropic's technology will provide the backend infrastructure, while OpenEvidence will optimize the system for each region based on local medical infrastructure. They did not disclose the financial terms of the partnership.

Anthropic unveiled its latest AI model, Claude Opus 5.5, on Tuesday, September 22, saying it performs on par with top-tier models such as Fable 5.1 while cutting the total cost of processing per task by as much as 40% compared with its predecessor, Opus 5. The launch comes amid international debate over the risks of advanced artificial intelligence, after Dario Amodei, Anthropic's chief executive, recently called on the global AI community to slow the pace of releasing new capabilities so that safety measures can keep up. Anthropic said Claude Opus 5.5 underwent external evaluation by independent AI safety research organizations Frontier Design and METR before release, and includes safeguards at the same level once reserved for the company's highest-tier systems. On software development benchmarks, the new model managed to outscore OpenAI's GPT-5.6 Sol, while its actual processing cost is only about one-third that of its competitor. API pricing stands at 4 US dollars per million input tokens and 20 US dollars per million output tokens, roughly 20% lower than Opus 5. It is already available through Amazon Web Services, Google Cloud and Microsoft Foundry on Azure. Internal evaluation data indicates that Opus 5.5 is up to 85% less likely to violate safety controls than Opus 5 or Mythos 5.1. Anthropic also plans to release sub-models Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks, and intends to expand a program granting vetted cybersecurity and life sciences researchers deeper access to the model.