Claude Sonnet 5 in Cursor: Pricing and Uses

Claude Sonnet 5Anthropic's medium-tier frontier model in Cursor, with a 200k standard context window and up to 1M context for longer tasks. Press Enter for the full definition. is Anthropic's current medium-tier model in Cursor, replacing Sonnet 4.6. Cursor rates it medium for speed and cost and frontier for intelligence. The model is meant to sit close to Opus 5 on real coding work while keeping the lower Sonnet price, which makes it a practical default when a task needs substantial reasoning but does not justify the most expensive Anthropic model.

Cursor highlights thinking mode, multi-step planning and debugging, and purposeful tool use. Those are selection signals rather than a promise that every request should use thinking. A small edit with a clear test may finish faster and more cheaply on ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. or another lower-cost model. Sonnet earns its place when the task has enough ambiguity or cross-file reasoning to use the extra depth.

Choose Sonnet 5 for everyday repository work that still needs frontier reasoning: planning a feature, tracing a multi-layer bug, reviewing an unfamiliar subsystem, or carrying a tool-driven change through tests. Cursor says Opus 5 remains stronger at peak quality on the hardest tasks. That makes the boundary about task difficulty and cost of failure, not a permanent preference for one model name.

A sensible escalation pattern starts Sonnet on a representative task with explicit acceptance checks. Move to Opus when Sonnet repeatedly misses a dependency, cannot hold the full constraint set, or produces a plan that fails before implementation begins. Stay on Sonnet when both models reach the same verified result, because Opus 5 lists rates of $5 input and $25 output against Sonnet's $2 and $10.

Do the comparison in fresh sessions with the same repository state. Switching models in one long thread changes the cache and leaves the second model reading the first model's attempted solution. A clean A/B task gives you a fairer view of final diff quality, elapsed time, redirects, and total usage.

Cursor lists a 200k context window and a 1M maximum. The larger ceiling can hold more repository material, tool results, and conversation history, but a ceiling is not a target. Loading an entire codebase can make the relevant file harder to distinguish and spends tokens on context the task may never use. Start with the smallest useful working set and expand only when evidence points outside it.

Sonnet 5 has an unusual pricing advantage for long context. Cursor states that all prompts bill at the base table rates even above 200k, with no separate long-context multiplier. That removes a price step, though it does not make the additional tokens free. A 600k prompt still contains three times as many input tokens as a 200k prompt before cache behavior is considered.

Sonnet 5 draws from the third-party Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool. Cursor lists $2 per million input tokens, $2.50 for cache writes, $0.20 for cache reads, and $10 for output. Other Models access starts with Pro, which includes $20 of monthly usage; Pro Plus and Ultra include more. Cursor StartAn India-only individual plan centered on Cursor Models, with cloud and mobile access but without Other Models, Auto, Bugbot, Automations or the Agent SDK. Press Enter for the full definition. includes only the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool, so it does not include Sonnet 5.

Output is the dominant published rate, so long explanations and large generated files can matter more than a short prompt suggests. Cache reads are one tenth of fresh input, which rewards stable repeated context. Changing models mid-session breaks the cache, so a workflow that hops between providers can cost more than four isolated rate cells imply.

Teams and Enterprise should also account for Cursor's Token Rate when they select a third-party model directly or when eligible Auto modes route to one. The rate is separate from Sonnet's provider pricing. Use the live usage dashboard for the effective account cost instead of multiplying only the model's input and output rows.

In Cursor, Sonnet 5 can search files and folders, search the web, read and edit files, run shell commands, control the browser, generate images, ask questions, and fetch rules. Tool availability still follows the product surface, approval mode, sandbox, connected credentials, and team policy. A model card listing a tool does not grant an account or bypass a repository control.

Roll it out with two or three task classes and a fixed review sheet. Record whether the result passed, how often a person redirected the run, elapsed time, token use, and which model would otherwise have handled the task. That evidence will tell you whether Sonnet belongs as a default, an escalation tier, or an option for one team whose work benefits from its reasoning style.

Keep the evaluation dated. Model behavior, Cursor's harness, and provider rates can change without the model name disappearing from an old report. A short rerun after a material release is more useful than treating one successful pilot as a permanent ranking.