Azure leak hints Claude Fable 5.5 is dropping next week

The most telling sign of an imminent frontier model drop is rarely a flashy keynote. It is usually an unannounced model slipping quietly into an enterprise cloud portal before anyone can stop it.

Anthropic's Claude Fable 5.5 just surfaced inside Microsoft Azure's East US 2 model catalog, briefly carrying a "Generally available" tag. While Microsoft and LiteLLM scrambled to pull the listing a day later, the leak instantly sparked reports that Anthropic could launch its next flagship as early as next week.

A cloud slip and a trail of pastry codenames

The leak broke when X user Alex Getman posted a screenshot showing Fable 5.5 listed on Azure. The entry, mirrored on LiteLLM, showed a 1 million token context window and pricing identical to Fable 5.1 at $10 per million input tokens and $50 per million output tokens. That said, those numbers look suspiciously like carried-over placeholders, especially after Azure queries began returning empty.

Yet Azure was hardly the only breadcrumb. Well-known leaker Lyra reported that partners had already secured access to the model, targeting a potential debut next week. Meanwhile, Lumina Bench spotted a codename dubbed "claude-onigiri-eap" on Artificial Analysis's frontend. It featured four effort levels and an Opus 4.8 fallback before disappearing hours later.

Other early access tags have surfaced across development tools. Cognition's Devin CLI logged orzata and vanilla, while user damageboy captured a Copilot picker screenshot showing "Claude Odango EAP" wedged right between Fable 5.1 and Sonnet 5.5. Across developer channels, consensus is leaning toward Onigiri being the internal name for Fable 5.5.

Self-improvement and the singularity call

Beyond raw benchmark muscle, what makes Fable 5.5 fascinating is the rumor that the model helped build itself.

Speaking at an IMF lecture in Bangkok, Anthropic co-founder Jack Clark boldly declared that "the singularity is here." Speculation has been running wild following hints from Anthropic that this model was substantially generated by self-improving AI agents that can churn out code around the clock.

That aligns with what developers noticed recently inside Claude Code and Opus 5.5, where heavy reasoning tasks were occasionally routed to a mysterious, unannounced checkpoint. If Anthropic's agents are now optimizing their own weights and pipelines, the iteration cycle is moving into unfamiliar territory.

Why it matters

We think the timing here is Anthropic's classic counter-punch play. Google unveiled Gemini 4 Argon on October 14, and dropping Fable 5.5 days later is clearly engineered to choke off Google's victory lap before it gathers steam.

For builders and enterprise teams, Fable models define the ceiling of agentic coding. Anthropic shipped Opus 5.5 less than a month ago to dominate high-end efficiency at lower pricing. Fable 5.5 is not about saving tokens; it is about pushing autonomous development into territory where human supervision becomes purely directional.

Anthropic's strategy looks razor-sharp, maintaining relentless cadence across low-latency and frontier tiers. But relying heavily on AI self-improvement introduces its own questions around subtle edge-case errors. If Fable 5.5 really lands next week, developers will quickly find out whether the singularity is ready for production.