SiteMax CMS

[{"h":"RUMOR: a second looped model may show up at DevDay","p":"In the early hours of September 24, the account @scaling01 posted that OpenAI will open what it called the forbidden axis at its September 29 DevDay: a second language model built on recurrent-depth architecture, alongside GPT-6 Astra. There is no name, no model ID, no price, and not a single line from OpenAI confirming the plan. It is a single-source rumor, but one with enough editorial weight to track, because it touches directly on the architectural thesis the company has been building for the past three weeks."},{"h":"Why recurrence matters","p":"Looped-depth models reuse the same computation block multiple times during inference instead of stacking new layers. In practice, that can raise reasoning capacity without multiplying parameters at the same rate. The catch is that the gain is not free: research on iso-depth scaling estimates that one extra loop delivers roughly 46% of what a genuinely new block would deliver. Real improvement, but with diminishing returns, which changes how any recurrent-architecture announcement should be read, not as a silver bullet."},{"h":"What OpenAI has said, and what it has not","p":"GPT-6 Astra, released September 3, is cited by outside analysts as the first production model from a major lab tied to this recurrent architecture, but OpenAI has never published a paper confirming the design. Chief scientist Jakub Pachocki said on September 2 that the computation-graph depth of frontier models, Astra included, sits within a factor of two of GPT-4, meaning bounded looping, not extreme recursion. Astra's system card also acknowledges that reasoning-trace monitorability dropped compared to GPT-5.6 Sol, a real admission that stands independent of which architecture caused it."},{"h":"Why the rumor is plausible","p":"OpenAI shipped GPT-6 Sol and GPT-6 Luna on September 22, building a price ladder on the same technical base as Astra, with cuts of up to 50% versus the previous generation. If a fourth model with recurrent architecture lands at DevDay, that suggests the company is treating looped depth as a permanent structural choice rather than a one-off experiment. That is a meaningful strategic shift, but one that still depends entirely on a confirmation that does not yet exist."},{"h":"Our read","p":"Worth tracking, not worth believing yet. A post with no model ID, no price and no second source is the kind of lead that can fizzle into nothing or become the story of the week within five days, since DevDay 2026 is set for September 29. If OpenAI confirms a second looped model, the practical effect is forced transparency around an architecture the company has so far only let researchers infer. Until then, treat this as a traceable hypothesis, not fact."},{"h":"Sources","p":"OrcaRouter, leak analysis based on @scaling01's post on X: https://www.orcarouter.ai/blog/openai-second-looped-model-devday-leak"}]