Frontier Model Launches Are Coming Faster

OpenAI, Anthropic and Google launched 14 frontier models between Jan 1 and Nov 7, 2026, up from 12 in all of 2025, by the dates of the labs' own launch posts. As of Nov 7, it had been 65 days since OpenAI's latest, GPT-6 Astra, 46 days since Anthropic's, Claude Opus 5.5, and 10 days since Google's, Gemini 4 Carbon. All three labs have had longer gaps earlier this year.

Anthropic launched the most

Anthropic launched five frontier models in 2025 and seven in 2026. Its gaps in 2025 ran 87, 75, 55 and 56 days, the last two around Claude Sonnet 4.5, which Anthropic called “our most capable model” on Sep 29, 2025. In 2026 its gaps ranged from 12 to 73 days, the shortest between Claude Opus 4.8 on May 28 and Claude Fable 5 on Jun 9. Since then it has alternated Opus and Fable models, with Opus 5 on Jul 24, Fable 5.1 on Sep 1 and Opus 5.5 on Sep 22.

OpenAI's count held as its gaps shortened

OpenAI launched four frontier models in each year. After GPT-4.5 on Feb 27, 2025, it waited 161 days for GPT-5, then 97 for GPT-5.1 and 29 for GPT-5.2. Its gaps in 2026 ran 84, 49, 77 and 56 days, the last from GPT-5.6 Sol on Jul 9 to GPT-6 Astra on Sep 3.

Google's launches come in pairs

Google launched three frontier models in each year, two of them close together each time. Gemini 2.5 Pro followed Gemini 2.0 Pro by 48 days in 2025, before a 238-day gap to Gemini 3 Pro. After Gemini 3.1 Pro on Feb 19, 2026, Google went 223 days to Gemini 4 Argon on Sep 30, then 28 days to Gemini 4 Carbon, which it called “our strongest model”. Google released both first to a group of cyber defenders.

Which models count

The count takes each lab's top-tier line, as the labs present it at launch: OpenAI's main GPT model, Anthropic's Opus and Fable models, and Google's Pro and Argon models. It adds any model a lab calls its most capable when it launches, such as Claude 3.7 Sonnet, Claude Sonnet 4.5 and Gemini 4 Carbon. It leaves out smaller tiers, OpenAI's o-series reasoning models, and restricted-access models such as Claude Mythos Preview and Mythos 5, and the Cyber models. It also leaves out Gemini 3.5 Flash, which Google called “our strongest agentic and coding model yet”, a claim limited to that work. Each model is dated by the post that introduced it. OpenAI previewed GPT-5.6 Sol to a limited group on Jun 26, 2026, and the count dates it by its general release on Jul 9. Anthropic's pages for Fable 5.1 and Opus 5.5 don't give a launch day: Fable 5.1 is dated by Anthropic's Enterprise Frontier Safeguards post of Sep 1, which names it, and Opus 5.5 by Anthropic's news page.