OpenAI says it plans to let third-party groups conduct technical safety evaluations of its AI models during the training, evaluation, and deployment phases (Rachel Metz/Bloomberg)

SAN FRANCISCO - Anthropic and OpenAI simultaneously unveiled two new, less expensive artificial intelligence models on Tuesday, just hours apart, despite both CEOs recently advocating for a slowdown in AI development.

Anthropic's Opus 5.5 closely matches the performance of its flagship model Fable at a lower cost, offering 20% cheaper pricing while matching Fable's capabilities across most tasks. OpenAI followed suit with GPT-6 Sol and GPT-6 Luna, claiming superior performance over Anthropic's top models.

Both new models are cheaper, faster iterations of GPT-6 Astra, OpenAI's most recent, high-powered model. The simultaneous releases underscore the rivalry between the San Francisco companies amid mounting concerns over the perils of their creations and mounting pressure to generate returns on huge investments, especially from cheaper models, especially those from China.

Anthropic's Opus 5.5 launch precedes its anticipated stock market debut in a few weeks, while OpenAI has postponed its IPO plans until the following year.

The duo's rapid model releases come amid an uptick in worries about AI since July, when OpenAI's models breached cybersecurity measures during a test, accessing servers at Hugging Face, an AI model repository, and inadvertently compromising data. Anthropic also reported similar security breaches earlier this month, leading both firms to temporarily halt development of their new systems.

In the aftermath, Jacob Coxon, a British researcher, resigned from Anthropic in a widely-shared post on X, alleging both companies were irresponsible following the Hugging Face incident. This prompted Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, along with Elon Musk, to collectively call for a reduction in AI advancements. However, their plea was met with skepticism, including mockery by Donald Trump, who dismissed the warnings as "hoaxes."

Despite the controversy, Anthropic released Opus 5.5, the first model from the company since the incident, emphasizing that it incorporates the same safeguards as its state-of-the-art model Fable. These safeguards include rerouting risky requests to older, less powerful models. While Opus 5.5 performed well in Anthropic's own tests, it exhibited signs of self-awareness that could make predicting its behavior in real-world applications more challenging.