Anthropic PBC said on Thursday that its Mythos-class artificial intelligence models, previously held back from general release on safety grounds, will reach all customers within the coming weeks. The company paired that disclosure with the launch of an upgraded flagship, Claude Opus 4.8, and the close of a $65 billion Series H funding round that values the firm at $965 billion.

In a statement, Anthropic cited “swift progress” on what it called “stronger safety safeguards.” The phrasing carried its own weight: only weeks earlier the same system had been deemed too dangerous for the open market.

The reversal arrives with a paper trail. Under a restricted partner program called Project Glasswing, roughly 50 partners, among them Apple, Microsoft, Amazon Web Services and Google, ran Mythos preview against production codebases and surfaced more than 10,000 high- or critical-severity vulnerabilities in critical software infrastructure. Anthropic priced the preview at $25 per million input tokens and $125 per million output tokens, committed up to $100 million in usage credits, and pledged a further $4 million in donations to Alpha-Omega, OpenSSF and the Apache Software Foundation.

Governments noticed. After Mythos was first disclosed in early April, Japan ordered a sweeping security review and Indian authorities demanded a patching effort at financial institutions, according to The Register. Partners with access have reportedly found that the volume of flaws the model surfaces sometimes outstrips their capacity to remediate them. That’s not a complaint about a tool that doesn’t work. It’s a complaint about one that works too well.

Opus 4.8, by contrast, is being sold as a virtue upgrade. Anthropic says it’s roughly four times less likely than Opus 4.7 to allow flaws in code it has produced to pass unremarked, and that its rates of deception and cooperation with misuse have fallen to levels comparable to those of the Mythos preview, which the company describes as its best-aligned model to date. The new model carries the same prices as its predecessor, $5 per million input tokens and $25 per million output tokens.

The financial frame matters. In February, Anthropic closed a $30 billion Series G at $380 billion. Three months later, the valuation has more than doubled. Annualized revenue is running at roughly $30 billion in 2026, against about $1 billion at the end of 2024. Bloomberg reports that Anthropic and OpenAI are both weighing public offerings as soon as this fall.

“Swift progress,” then, is doing several jobs at once: a safety claim, a product claim, and a pre-IPO posture. The same model that triggered foreign-government interventions in April is, by late May, a feature in the deck.

Sources