Where the Two Companies' Cadence Differs
Heading into mid-July, the release cadences of the two leading labs show a clear contrast. On OpenAI's side, the GPT-5.6 previewed June 26 in three tiers (Sol/Terra/Luna) is gradually rolling out at volume: the middle tier Terra opened an enterprise waitlist, the fast tier Luna is in small-scale testing for developers, and the flagship Sol remains gated but is on the whole being pushed toward "available." On Anthropic's side, what's broadly available is Fable 5, while the strongest Mythos 5 remains mostly limited to approved institutions and critical-infrastructure defenders, and its public timeline for broader users has been pushed back several times. All of the above progress is per the two companies' official disclosures.
The Main Cause of the "Delay" Is Review, Not an Inability to Build It
To be clear, Anthropic's delay, per public information, isn't mainly an inability to build it technically, but the result of external review compounded with its own safety process: Mythos 5 was previously required to be taken offline over export controls, and only obtained partial clearance—for use by defenders first—after weeks of negotiation with the responsible agency; broader public access must continue to clear the frontier-model review gate, and combined with Anthropic's consistent practice of placing safety evaluation before release, the timeline has kept slipping. In other words, this looks more like a "compliance cadence" than a "capability gap"; the specific reasons and timing are per the official word, and some steps remain unconfirmed.
What It Means for Users and Developers
Putting the two threads together, the real change isn't whose model is stronger, but that "when the strongest model launches, and to whom" is increasingly decided by review and compliance. For those selecting, the pragmatic move is: don't bet that some top model will "go public soon," but plan around the versions you can actually get and use compliantly right now—on the OpenAI side, look at the tiers that can be deployed; on the Anthropic side, focus on the broadly available Fable 5, and treat restricted top-end models like Mythos 5 as "icing on the cake, deal with it once you have it." The time gap between frontier capability and "capability you can actually use" will persist in the short term.
via: Synthesis of OpenAI News and Anthropic News official disclosures