Frontier AI keeps racing despite calls to slow down

Despite a July open letter from AI employees urging frontier AI labs to decelerate their development pace, major model releases continue to proliferate swiftly. Anthropic and OpenAI each unveiled fresh versions of their Claude and GPT models on Tuesday: Opus 5.5 and GPT-6 Sol and Luna. CEO Dario Amodei of Anthropic reiterated the staff's message in a September 12, 2026 post, emphasizing the need to moderate AI model improvements.

Ten days later, Anthropic announced Opus 5.5, touting it as the most powerful model they have tested. The release follows a rapid succession, with model updates every few weeks in 2026, compared to a quarterly cadence in 2025. OpenAI's release schedule has also accelerated, with multiple model variants in February, March, April, and subsequent releases in July and September 2026.

Both companies are simultaneously preparing for initial public offerings. Opus 5.5 reportedly outperforms its predecessor in software tasks, enabling swift code migrations and reduced software load times. Artificial Analysis evaluates Opus 5.5 as equivalent to GPT-6 Astra for agentic work. The Claude model has topped the Artificial Analysis Intelligence Index with a score of 58, based on performance in 10 benchmark evaluations.

Despite its superior performance, Opus 5.5 is reportedly 40% cheaper than its predecessor, with lower costs for input and output tokens, cache reads, and faster output generation. OpenAI also announced lower costs for GPT-6 Sol and Luna, leaving high-performance workloads to its Astra model. Both companies tout improvements in alignment and safety for their models, though Anthropic's defenses are partly aimed at safeguarding its business from competition.