Anthropic is preparing for the release of the Claude Sonnet 5.5 model, which is currently undergoing beta testing within Claude Code. The model's identifier, 'claude-sonnet-5-5', has been detected in configuration files, alongside references to 'claude-opus-5-5' and 'claude-fable-5.1'.
Early testing suggests that Sonnet 5.5 offers performance levels approaching the flagship GPT-6 Astra, while significantly outperforming GPT-6 Sol in coding tasks and agent capabilities. Reports indicate that the model is priced at $2 per million input tokens.
Developers have noted that Sonnet 5.5 demonstrates high efficiency in code generation. In comparative tests, the model produced files ranging from 90 KB to 131 KB, while GPT-6 Sol generated files between 20 KB and 29 KB under similar conditions.
Anthropic has reportedly integrated AI into its own development process, with 26% of core R&D tasks now handled by previous Claude models. The company aims to achieve full automation in model development by 2027 through recursive self-improvement.
The release of Sonnet 5.5 is expected to coincide with OpenAI's DevDay, scheduled for September 29. Anthropic previously launched Opus 5.5 on September 22, achieving a 40% reduction in costs and a 30% increase in speed compared to the previous generation.
