Claude Sonnet 5.5 lands in Cursor with Opus-class coding scores

Cursor has added Claude Sonnet 5.5, a mid-tier model that Anthropic says approaches Opus 5.5 on many tasks. The published results are strongest in agentic coding, where Sonnet surpasses the flagship on one terminal benchmark while using fewer tokens and tool calls.

Sonnet 5.5 is the second model in Anthropic’s Claude 5.5 family, arriving one week after Opus 5.5. Opus targets ambiguous, open-ended work that requires sustained judgment; Sonnet handles routine coding and knowledge work at lower cost.

• Speed: More than 30% faster output than Sonnet 5, according to Anthropic.
• Task cost: Up to 30% lower through reduced token use and fewer tool calls.
• API price: Unchanged from Sonnet 5.
• Cursor use cases: Bug fixes, refactors, scoped features, and tool-heavy agent runs.

Fewer agent steps cut the bill

Anthropic attributes the lower per-task cost to behavioral efficiency because the API rates remain unchanged. A tool call occurs whenever an agent invokes an external capability, such as searching a repository, reading a file, running a command, or editing code. Fewer calls reduce latency, token consumption, and the risk of an agent drifting during a long task.

Lovable co-founder and CTO Fabian Hedin said the company’s evaluations found that Sonnet 5.5 used about one-third fewer tool calls and roughly half as many shell executions to complete coding jobs, VentureBeat reported. Teams running agents inside Cursor or through the API should see those reductions in completion time and request budgets, though results will vary with prompts, tools, and repository size.

Sonnet edges Opus in the terminal

Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, which measures how well an agent completes practical tasks through a command-line environment. Anthropic reports 66.4% for Opus 5.5 and 10.3% for Sonnet 5 on the same benchmark.

On GDPval-AA, a benchmark for economically valuable knowledge work, Sonnet 5.5 nearly matches Opus 5.5. Anthropic still gives Opus the advantage on complex assignments that require long-horizon planning and judgment. Benchmark outcomes depend on prompts, agent scaffolding, tool permissions, and scoring methods, so production evaluations should use representative codebases and tasks.

A practical split for model routing

Sonnet 5.5 fits well-scoped work such as fixing bugs, refactoring modules, implementing small features, and generating documents, slides, or spreadsheets. Its lower tool usage also suits repeated agent runs where latency and request volume affect operating costs.

Opus 5.5 remains the stronger choice for architectural decisions, ambiguous investigations, and long tasks that require the model to preserve context and judgment across many steps. Teams can route routine work to Sonnet and reserve Opus for cases where additional reasoning capacity justifies the higher token price.

One API setting needs attention

Claude Sonnet 5.5 is available through the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. Anthropic offers zero data retention for the model, and direct API clients can select it with the model ID claude-sonnet-5-5.

API integrations that currently run Sonnet with thinking disabled must adopt the new between_tools setting before migrating. Opus 5.5 already rejects requests that disable thinking entirely. Cursor manages this request configuration inside the IDE, while developers maintaining custom agents should update and test their configuration before changing the production model ID.