CursorBench 4.0

Every result

• Claude Opus 5.557.8%Reported by the model's maker (Anthropic), Sep 22, 2026

  Anthropic reports 57.8% CursorBench 4.0 accuracy for Claude Opus 5.5 at max effort, using production safeguards and possible fallback routing.

  Tasks: Ambiguous multi-file coding tasks from real Cursor sessions; exact split and task subset unreported.

  Setting: Adaptive thinking at max effort: the Opus announcement applies max effort unless otherwise noted; its Terminal-Bench exception does not apply to this row. Numeric reasoning/token budget unknown.

  Limits: Provider-reported result, not an independent run, private test, overall ranking or matched-budget comparison. No CursorBench uncertainty interval or standard error was established. The Terminal-Bench ±2.6 percentage-point standard error does not describe this CursorBench result. The documented serving configuration can include Opus 4.8 cybersecurity and Opus 5 biology/frontier-development fallbacks. Whether any fallback ran on this CursorBench row is unestablished; no pure-weights claim. Exact task split, sample size, trials, pass@k, aggregation, harness version, prompts, tool budget and execution date remain unknown. Full evaluation methods remain unverified; the announcement gives only partial methodological detail. The API default medium effort differs from this max-effort setting. Sonnet 5.5 reports 55.5% on the same named benchmark/version, but its effort and matched evaluation configuration are unknown.

  Anthropic · Anthropic · Anthropic · Anthropic

• Claude Sonnet 5.555.5%Reported by the model's maker (Anthropic), Sep 28, 2026

  Anthropic reports 55.5% accuracy for Claude Sonnet 5.5 on CursorBench 4.0. This is a provider-reported observation; benchmark effort and full evaluation methods remain unknown.

  Tasks: Ambiguous multi-file coding tasks from real Cursor sessions; exact split and task subset unreported.

  Limits: Provider report only; no independent run, private test, overall ranking or matched-budget comparison. The announcement does not identify Sonnet effort for this result. API high and app Medium defaults do not establish the benchmark setting. Exact task subset, sample size, trials/pass@k, harness, prompts, tool/reasoning budgets, aggregation, execution date and uncertainty remain unknown. Announcement describes higher-risk cybersecurity fallback to Sonnet 5 at launch; whether safeguards were enabled or any fallback triggered on this benchmark is unknown. No isolated fixed-weight claim. Full evaluation methods remain unverified; the announcement gives only partial methodological detail. Opus 5.5 has a separately reported CursorBench 4.0 result of 57.8% using max effort. Sonnet effort and matched harness/budget/configuration are unestablished, so no quality ranking or meaningful 2.3-point advantage is claimed.

  Anthropic · Anthropic · Anthropic · Anthropic

• Claude Opus 546.6%Reported by the model's maker (Anthropic), Sep 22, 2026

  Anthropic reports 46.6% for Claude Opus 5 on CursorBench 4.0 at max effort. The result is provider-reported; missing evaluation details limit comparison.

  Setting: Max effort; numeric reasoning and token budgets are not specified.

  Limits: Anthropic reports this result; it has not been independently reproduced here. The evaluated checkpoint and serving configuration are not identified. This result refers to the named older release, whose API identifier and current availability are unknown. The task subset, sample size, harness, prompts, run date, aggregation and confidence interval are not reported in the inspected narrative. Tool use and numeric tool, runtime and token budgets are unspecified. Max effort is reported, but the numeric reasoning budget is unspecified. Results with the same benchmark name can be viewed together as reported scores. Unmatched settings and missing methods prevent a controlled quality ranking or a conclusion that small differences are meaningful.

  Anthropic

• Claude Sonnet 534.1%Reported by the model's maker (Anthropic), Sep 28, 2026

  Anthropic reports 34.1% for Claude Sonnet 5 on CursorBench 4.0. The result is provider-reported; missing evaluation details limit comparison.

  Limits: Anthropic reports this result; it has not been independently reproduced here. The evaluated checkpoint and serving configuration are not identified. This result refers to the named older release, whose API identifier and current availability are unknown. The task subset, sample size, harness, prompts, run date, aggregation and confidence interval are not reported in the inspected narrative. Tool use and numeric tool, runtime and token budgets are unspecified. Reasoning effort is unspecified; settings from other releases or evaluations cannot fill that gap. Results with the same benchmark name can be viewed together as reported scores. Unmatched settings and missing methods prevent a controlled quality ranking or a conclusion that small differences are meaningful.

  Anthropic