Anthropic Testing Claude Opus 5.1 & Sonnet 5.1

Anthropic may be testing its next-generation Claude models under the codenames "Marshmallow" and "Melon." Leaks suggest these models are internally identified as claude-marshmallow-eap and claude-melon-eap. Here's the thing: Marshmallow reportedly performs better than Melon and even outperforms the current Opus 5. This has led to speculation that Marshmallow could be Claude Opus 5.1, while Melon might be Claude Sonnet 5.1. Both are said to be in early-access testing. What's interesting is that neither Marshmallow nor Melon reportedly reaches "Fable-level" capabilities, suggesting they might be evolutionary updates rather than a major leap. If these are indeed Opus 5.1 and Sonnet 5.1, users could see improvements in reasoning, coding, and reliability. This development could mean Anthropic is refining its lineup to offer models for different user needs.