OpenAI and Anthropic's new flagship models revealed on the same day
OpenAI is gray-testing GPT-6.2 Astra in the Codex backend, and Claude Mythos 5.5 was discovered in Anthropic's source code; leaks about the two next-generation models appeared on the same day. The leaked billing shows that gpt-6.2-astra already accounts for 19.8% of real traffic, while the older gpt-6-astra accounts for 80%. The article states that Mythos 5.5 and Fable 5.5 share the same underlying foundation and identical API pricing; Mythos 5.5 is a research version with the standard refusal limits lifted, has appeared in the official MLflow and Azure catalogs, and supports a 1 million Token input context and a maximum output of 128,000 Tokens.
Why read this
AI score 60
Two flagship models from the two companies leaked on the same weekend, and the article provides the gray-test traffic share and version differences, which can be compared to see the competitive rhythm of the next-generation models.
Topics
Body · AI translation
OpenAI and Claude's New Flagships Exposed on the Same Day Xinzhiyuan November 02, 2026 07:07:36 From Beijing
Two major new model leaks actually landed on the same day!
Just days after Google's Gemini 4 Carbon snatched back the global number one spot, OpenAI and Anthropic simultaneously revealed their next-generation trump cards over the weekend.
OpenAI skipped straight past 6.1 and quietly ran a gray-scale test of GPT-6.2 Astra in the Codex backend.
No sooner had this leak shot to the top of trending searches than Claude Mythos 5.5 was dug up lying plainly in Anthropic's official source code.
In the account billing details posted by Token Gremlin, the older gpt-6-astra accounted for 80% of calls, while the separately listed gpt-6.2-astra had already taken in 19.8% of real traffic.
First, it fixed 6.1's "unauthorized concealment" security red line.
Second, it cured the first-generation 6.0's lopsided flaw of being "high-IQ but unable to do the work."
Codex head Tibo himself personally showed up in the comments to respond, outright spoiling: "I strongly demand we reset the countdown and go another round!"
In fact, Mythos 5.5 and Fable 5.5 use the exact same AI brain underneath, and even their API pricing is identical down to the cent.
Among them, Fable 5.5 is equipped with strict safety guardrails to intercept high-risk instructions, and is fully open to developers worldwide as the general-purpose commercial flagship.
The other version, Mythos 5.5, is a "special research edition" with the regular refusal limits unlocked.
At present, it has already appeared simultaneously in the official MLflow and Azure catalogs, confirming support for an input context of up to 1 million tokens and a maximum output of 128,000 tokens.
Silicon Valley's three giants collided in early November 2026 into the most densely packed week of ace-model showdowns in history.
The real show is finally about to begin.