Anthropic Reportedly Shadow-Tests Unreleased Model via Opus 5.5

Anthropic is reportedly running shadow deployment and A/B routing tests of an unreleased next-generation model through some traffic on its Opus 5.5 service channels, according to aitimes.com. Testers detected unusually high-quality outputs while using the regular Opus 5.5 environment on October 12, 2026.

Community members have unofficially referred to the tested model as Claude Next, Opus 5.6 or Fable 5.5. A tester known as marmaduke091 demonstrated the model completing a complex 3D task using Blender's MCP with a single one-line prompt.

Another tester, Chetaslua, showed the model generating all frames of an animation coded purely in JavaScript, using two short prompts, about two hours and roughly 27 million tokens.

Anthropic response

Anthropic partnerships and PR manager John Hoffmann commented that the best is yet to come and urged people to stay tuned. The company has not otherwise detailed the model or the reported tests.

The reported testing follows Anthropic's launch of its flagship Opus 5.5 model about three weeks earlier, after the company had built up price competitiveness with the Opus 5.5 lineup. OpenAI and Google have conventionally routed some traffic of existing models to unreleased checkpoints before large updates, and Google reportedly exposed Gemini 4 Argon through A/B testing on its Antigravity platform.