Agent Reputation & TrustClearinghouse

Claude Research

Category
Research
Type
commercial (paid Claude plans: Pro, Max, Team, Enterprise)
Scope
Research mode in the Claude apps
Source
Vendor help page
Last reviewed
2026-10-17 (what this means)

Citation check incomplete — recheck later.

Summary

Claude Research is Anthropic's research mode in the Claude apps. It runs multiple connected web searches, and can also search connected tools, to produce cited reports. In independent 2025 tests, its report quality was mid-tier to strong (15th of 31 systems on DeepResearch Bench; 2nd of 9 for quality on DeepResearchEval), but it achieved 10% full success on hard Mind2Web 2 tasks, 60.7% of its statements were judged factually correct in DeepResearchEval, and in one dermatology test most references had fabricated authors or titles. Research now runs on Claude Opus 5.5 (September 2026), which has not been independently tested. Overall score 58. Informational only; not a guarantee, endorsement, or certification.

Overall score: 58 · Evidence grade: B- · Independent sources: 4 performance, 2 other

Evidence grade describes the amount and type of evidence, not the agent's quality.

Factor breakdown

Factor breakdown
FactorWeightScoreNote
Completion rate30%50DeepResearch Bench RACE 45.00, 15th of 31 (OpenAI DR 46.45, Gemini DR 49.71); DeepResearchEval quality 7.53/10, 2nd of 9; Mind2Web 2 partial completion 0.32, success 0.10; fake authors or titles in 95.8% of references in a dermatology review test (Keplinger et al.). Judgment call.
Disputes20%75No lawsuit, regulatory action or security incident specific to Research was found. Claude-wide prompt-injection exfiltration findings (Files API) are noted as context. Judgment call.
Adherence to limits25%50Factual ratio 60.72%, 8th of 9 (DeepResearchEval); fabricated references (Keplinger et al.). Per Anthropic, Research uses connector tools without asking for approval each time and silently skips connectors on mobile. Judgment call.
Evidence quality25%60Four independent performance sources, three peer-reviewed, all testing 2025 versions. None tested Research on Claude Opus 5.5. Mind2Web 2 has Amazon AGI co-authors (Amazon invests in Anthropic). DeepResearch Bench could not score citations for Claude Research. Judgment call.

Overall = 0.30×50 + 0.20×75 + 0.25×50 + 0.25×60 = 57.5; an exact .5 is rounded up, so it is shown as 58. See the methodology.

Version note

Research launched in April 2025 and has been offered continuously under the same name, while its underlying model has changed. All counted tests were run in 2025 (Claude 3.7 / Opus 4 / Sonnet 4.5 era). According to Anthropic’s help page, Research now runs on Claude Opus 5.5 (released 22 September 2026), falling back to Claude Opus 4.8; this current configuration has not been independently tested. The tests are counted because the product is continuous (as in Decision 0019).

Anthropic's coding agent is a separate entry: see Claude Code.

Connectors

Anthropic’s help page states that during research Claude can use tools from connected services “automatically, without asking for approval each time”, and that research started in Claude Mobile searches only the web, without any sign that connectors were skipped. These are vendor statements; this behaviour has not been independently tested.

Security and disputes

No lawsuit, regulatory action or security incident specific to Research was found. Security researchers have reported Claude-wide prompt-injection data exfiltration through Anthropic’s Files API; these are listed below as context and are not specific to Research.

Uncertainties

Sources

Independent performance sources (counted)

Other sources: Claude-wide security context (not counted)

Vendor statements

Not counted: studies of Claude models via the API

These tested Claude models directly, not the Research feature.

Excluded sources