Another day, another frontier AI model announcement—actually two of them, one from Anthropic and one from OpenAI. It feels like new model releases are accelerating, and they are, but many of the new releases are models that repackage (and often reprice) the capabilities of earlier flagship models.
Both of Tuesday’s releases are examples. OpenAI announced GPT-6 Sol and GPT-6 Luna—faster, more affordable models that bring the advances behind its flagship GPT-6 Astra “to everyday work.” Anthropic announced Claude Opus 5.5, which it says matches the performance of its flagship Claude Fable 5.1 model on most work and costs around 40% less to run than its predecessor, Claude Opus 5. (On Monday, SpaceXAI launched its Grok 4.7 model, which the company says excels at coding and knowledge work and can run longer tasks and verify its outputs. But Grok 4.7 is a legitimately new flagship model with new performance and capabilities, not an optimized version of an earlier model.)
Anthropic’s release cadence for frontier models has roughly doubled during 2026, from a model every 46 days in the first half to every 26 days so far in the second; OpenAI’s release cadence has increased from every 46 days in the first half of the year to every 51 days so far in the second. But the time between genuinely new flagship frontier models hasn’t changed much during 2026 for either company, the release dates show. This year, Anthropic has released eight flagship frontier models; OpenAI has released six.
A faster release cadence doesn’t necessarily mean the underlying technology is advancing faster. Increasingly, one major model breakthrough can spawn a family of cheaper or more specialized versions aimed at different customers, which makes the pace of product launches a less reliable proxy for the pace of progress.
The AI labs say they are increasingly using AI itself to design and build new models—a practice called “recursive self-improvement.” Anthropic reports that as of August, its Claude models were leading 26% of its AI R&D work and collaborating on more than 90%.
Researchers are beginning to use AI models to design new computing infrastructure that delivers more computing power, efficiency, and speed. They’re using AI to generate synthetic training data and to optimize the whole software framework that manages model training. And they’re using AI coding models to write and optimize the code that defines and implements the model.
OpenAI says that before June 2026, its total AI agent run time was less than human-labor hours, but as of mid-August, its AI agents were working 3.1 days for every 1 day of human labor (based on an eight-hour workday).
Arnal Dayaratna, research vice president of software development at IDC, questions whether the labs have actually begun to see the effects of real self-improving systems. His take is that any acceleration in model releases is driven by the AI industry as a whole gaining a better understanding of what the market wants, as well as by competitive pressure from open model developers.
Model release dates can be driven by a number of different factors—not just speed of innovation, Gartner AI analyst Arun Chandrasekaran points out. “AI labs time releases to defend market share, lock in enterprise customers, or shape investor expectations,” he says in an email to Fast Company. “The frequent releases are a signal that demonstrates agile development, and provide evidence of improving model price/performance.”
OpenAI and Anthropic are both working toward an IPO within the next year, and the gravity of that moment can influence model release schedules and marketing pushes. The labs have said a lot to convince investors and the public that they have a reliable formula (aka scaling laws) for systematically improving their AI models until they become as capable as humans at most tasks and, later, far more capable than humans at almost everything.
“Ahead of a listing, OpenAI and Anthropic need to prove that billions of dollars in compute and research spending are producing a repeatable product engine, not a single breakthrough,” says PitchBook senior analyst Harrison Rolfes in an email to Fast Company. “The harder question for IPO investors is whether faster releases produce durable revenue and margins, or simply shorten each model’s shelf life while keeping compute, safety, and infrastructure costs elevated.”