There used to be a simple pattern for a new frontier model: announce it, publish a benchmark table, open the API. In 2026 all three of the biggest US labs broke that pattern in the same way. Their most capable models went first to cyber defenders.

Anthropic: Mythos and Fable

Anthropic unveiled Mythos in April and restricted it to selected organisations through Project Glasswing because of its cyber capabilities. On June 9 it released Claude Fable 5, a safeguarded public version with classifiers that block high-risk cyber and biological requests, priced at $10 per million input tokens and $50 per million output tokens, according to CNBC.

On November 9 Anthropic launched Fable 5.2 and Mythos 5.2, which it says are 30% faster and 20% cheaper, at $8 and $40 per million tokens. Fable 5.2 is generally available; Mythos 5.2 remains limited to vetted organisations.

OpenAI: GPT-6

OpenAI released GPT-6 Astra on September 3, first to companies in its application-based Daybreak cybersecurity programme, CNBC reported. OpenAI said it was the first model to reach the Critical level of its internal cybersecurity framework, and Sam Altman said it had been through a formal review with the US administration before release.

On October 7, GPT-6 Sol (for paid plans) and GPT-6 Luna (for free users) replaced GPT-5.6 in ChatGPT.

Google: Gemini 4

Google followed the same path. Gemini 4 Argon rolled out on September 30, first to defenders through its Fairwind programme. Google says Argon can find, validate and patch critical software vulnerabilities on its own, and reports a score of 77.9% on the DeepSWE v1.1 coding benchmark. Introductory pricing is $2 in and $10 out per million tokens, rising later to $4 and $20.

Gemini 4 Carbon, Google’s strongest model, followed on November 10, with its introductory price ending on November 30. Gemini 4 Hydrogen, the fast and inexpensive tier at $1.50 in and $7.50 out with a one-million-token input window, launched on November 12 across the API, AI Studio, the Gemini app and Search.

Why it matters for developers

  • The top tier comes with a queue. The most capable model from each lab now often requires an application, a vetting process or an enterprise agreement before it reaches the general API.
  • Government review is part of launch. At least one release went through a formal review with the administration first, and Anthropic’s models were briefly frozen by an export directive in June. Launch dates are less predictable than they were.
  • Pricing diverged. Anthropic cut prices with its 5.2 release, while Google’s introductory Gemini 4 pricing doubles after the launch window. Budget for the post-launch price, not the intro one.
  • Cheap tiers matter more. Models like Hydrogen show the labs competing hard on cost per task, which is what most production apps actually pay for.

All benchmark figures are as reported by the vendors.

Sources

  1. CNBC: OpenAI Astra GPT-6 and cyber (Sep 3, 2026)
  2. Google: Gemini 4 Argon (Sep 30, 2026)
  3. Google: Introducing Gemini 4 Hydrogen (Nov 12, 2026)
  4. CNBC: Anthropic launches Claude Fable 5.2 and Mythos 5.2 (Nov 9, 2026)
  5. Anthropic: Claude Mythos

We link every claim to the reporting or official statement it comes from. Where a detail rests on one outlet’s reporting, we say so. Spotted an error? Reply to any newsletter and we’ll correct it.