Anthropic could drop Opus 5 any moment, as GPT 5.6 challenges Claude's dominance in AI coding

Claude Opus 5 appears to have leaked through Anthropic’s own fallback system after a Fable 5 safeguard routed a conversation to “Opus 5 High,” according to a screenshot circulating on X and in AI developer circles.

The screenshot shows a Claude conversation where the user asked for research ideas beyond the usual math conjecture trend.

The response discusses algorithm discovery, quantum circuit synthesis, numerical analysis and clinical trial reanalysis, but the important part is at the bottom of the screen: “Switched to Opus 5.”

Opus 5

The model picker also shows “Opus 5 High,” while another option says “Edit and retry with Fable 5.”

A second screenshot of a post attributed to an Anthropic employee appears to confirm what happened: Fable 5’s sensitive biology classifiers were triggered because the prompt spread across several research fields, so Claude “dropped” the session down to another model.

Fable 5 was supposed to fall back to Opus 4.8

This is notable because Anthropic’s public Fable 5 setup uses fallback routing for some sensitive areas.

When Fable 5’s classifiers detect certain requests in fields such as cybersecurity, biology, chemistry or model distillation, Anthropic says the request can be handled by Opus 4.8 instead of Fable 5.

If the screenshot is real, Anthropic may already be testing Opus 5 as the new fallback model for some Fable 5 guardrail cases.

That would make sense if Opus 5 is close to release, especially because Fable 5 has been powerful but difficult to manage across safety, cost and usage limits.

Anthropic has not publicly announced Opus 5 yet, so the screenshot should still be treated as an early signal rather than confirmation of a launch.

It does, however, match the recent pattern of users seeing strange routing behavior, with some claiming they were moved from Opus 4.8 to a noticeably stronger model.

Opus 5 was expected this week, but the launch slipped

Opus 5 was expected to launch this week, but the release appears to have been delayed at the last minute. Some early users who believe they were routed to the new model have reported stronger visual and coding output, including detailed 3D SVG generation.

That kind of staged testing would not be unusual. Anthropic can quietly route a small number of real sessions through a new model, check output quality, measure cost, and validate safety behavior before opening access more widely.

The delay may also be tied to cost. Opus 5 is reportedly very token-hungry, and that matters because Anthropic is already managing Fable 5 access carefully.

Powerful coding and research models can burn through context quickly, especially when they are used inside Claude Code or long research sessions.

That pressure is part of the wider AI compute spending problem now shaping every major model launch.