Since Anthropic announced Opus 5.5 on September 22, it also teased that "Sonnet 5.5 and Haiku 5.5 will be released in the coming weeks." Sonnet 5.5 has arrived much faster than expected. Starting from last night, rumors about its upcoming release spread on the X platform, with multiple developers stating that the model has entered its final stage before release and has already begun a gray-scale test in Claude Code, or been routed to the 5.5 version for "backdoor testing."
Exclusive Identifier Appears, Gray-Test Users Get Early Access
Developer leaks show that the artifact has already featured the exclusive model identifier for Sonnet 5.5, "claude-sonnet-5-5." The same configuration also lists "claude-opus-5-5," "claude-sonnet-5," and the mysterious "claude-fable-5.1," indicating that the model is already implemented and just waiting for a switch. Netizens summarized a verification method: after turning off internet and memory in Claude Code, ask "Do you know Tibo, the reset guy?" Accounts routed to 5.5 can accurately tell the Codex reset joke. Some people have quietly experienced it ahead of time.
Real Test Beats Sol, Price Only $2 per Million Tokens
Shared tests show that Sonnet 5.5 is closer to Opus 5.5 but much cheaper. In a pure JS frontend UI/animation showdown, developers said it "beats OpenAI's Sol and Astra badly." @chetaslua ran the same three prompts six times each, and Sonnet 5.5 generated files of 90–131KB, often hitting the 90-minute limit. GPT-6 Sol only produced 20–29KB and took 9–12 minutes. He said that in Ultracode mode, Sonnet 5.5 "got back the human touch of the Sonnet 4 era." There are even leaks suggesting that its advanced coding and Agent capabilities have already caught up with the flagship Astra.
The pricing is a "price killer": input at $2 per million tokens, output at $10 per million tokens, and cache reading only $0.20 per million tokens. It directly enters the price range of OpenAI's GPT-6 Sol/Luna but with better performance. Developer reviews say if Sonnet 5.5 is truly close to Opus 5.5, Anthropic has made cutting-edge capabilities extremely cheap, "this is what OpenAI should be most worried about."
Directly Attacking DevDay, Behind Is the AI-Generated AI Flywheel
The timing of the release is dramatic: Tuesday this week (September 29) is OpenAI's annual developer conference, DevDay. Multiple sources indicate that Sonnet 5.5 is expected to be released next Tuesday, directly attacking the competitor. Industry speculation suggests that the core reason for Anthropic's rapid iteration is a high-standard "data flywheel"—exposure data shows that 26% of core R&D work is already led by the previous generation of Claude, and more than 80% of self-generated code is written by large models themselves. Recursive self-improvement (RSI) is beginning to show, and the company expects Claude to be fully automated as early as 2027. The emergence of Sonnet 5.5 is seen as a signal that the era of affordable, high-quality AI has officially arrived.
