Anthropic Releases Claude Sonnet 5.5 with Faster Speed and Lower Cost
Anthropic has released Claude Sonnet 5.5, which runs over 30% faster and costs up to 30% less to execute than its predecessor. Additionally, Anthropic has upgraded its free tier on Claude.ai to use Sonnet 5.5 and confirmed that Haiku 5.5 will launch in the coming weeks. This update significantly lowers the cost barrier for complex AI applications, outperforming even the larger Opus 5.5 on certain agentic coding tasks by allowing users to run parallel agents within budget. Offering a highly capable model on Claude.ai's free tier also sharpens Anthropic's competitive edge against rival platforms like OpenAI's ChatGPT. Sonnet 5.5 incorporates upgraded cybersecurity safeguards equivalent to Anthropic's Opus-level protection protocols. Early hands-on testing revealed an issue with extended reasoning where setting thinking effort to 'max' caused the model to exhaust a 128,000-token limit without generating output, though slightly lower reasoning settings succeeded.
## BACKGROUND
Modern large reasoning models utilize extended reasoning modes, generating internal 'thinking tokens' before outputting a final response to solve multi-step coding or math problems. Because these reasoning tokens are billed at higher output rates, allocating and managing token budgets is essential to prevent models from over-thinking and draining compute resources.