Kimi Researcher
- Category
- Research
- Type
- commercial (Kimi membership credits)
- Scope
- Deep Research mode in Kimi (Moonshot AI)
- Source
- Vendor help page
- Last reviewed
- 2026-10-19 (what this means)
Citation check incomplete — recheck later.
Summary
Kimi Researcher is Moonshot AI's deep-research agent, offered as Deep Research in Kimi. Two independent evaluations were found. On DeepResearch Bench (mid-2025) its report quality was mid-tier, close to Claude Research and below OpenAI and Gemini deep research. On ICBCBench, a 2026 financial-research preprint, it ranked third among closed systems. Fewer than three independent sources were found, so no overall score is given. In April 2026 the Kimi app reportedly showed one user's résumé to another user. Insufficient evidence. Informational only; not a guarantee, endorsement, or certification.
Status: Insufficient evidence — no overall score is given (fewer than three independent performance sources). Independent performance sources: 2.
Factor notes (no scores)
| Factor | Weight | Score | Note |
|---|---|---|---|
| Completion rate | 30% | — | DeepResearch Bench RACE 44.64 (OpenAI DR 46.45, Claude Research 45.00); ICBCBench English overall 47.59 (Gemini DR 57.38, o3-deep-research 54.67), Chinese 44.72. DeepResearch Bench gave no citation-accuracy (FACT) score because Kimi's citations could not be parsed from the interface. A later re-scoring of the same reports by Microsoft authors (DualGraph preprint, February 2026, GPT-4.1 judge) reported 20.80% citation accuracy with 2.30 effective citations per task, the lowest of the systems listed; not counted (same reports, no new test), and the figure may reflect the same parsing problem rather than citation quality. |
| Disputes | 20% | — | April 2026: the Kimi app reportedly returned another user's real résumé during an unrelated request (Kimi-wide, not specific to Deep Research); no official statement. No regulator action against Kimi found. |
| Adherence to limits | 25% | — | Not independently tested. |
| Evidence quality | 25% | — | Two independent performance sources (one peer-reviewed, one preprint with Alibaba co-authors). Vendor benchmark claims (HLE, xbench) excluded. |
See the methodology for the insufficient-evidence rule.
Version note
Kimi-Researcher was announced on 20 June 2025 and is offered as Deep Research in Kimi. Moonshot’s help page says Deep Research is powered by the Kimi-Researcher model. Kimi’s general models have since changed (K2.5, K2.6, K3), but no change of the Deep Research model has been announced, so the product is treated as continuous.
Disputes
On 20 April 2026 the Kimi app reportedly returned another user’s real résumé (name, phone number, email and work history) during an unrelated translation request. Staff reportedly described it first as an “AI hallucination” and then as a cross-user mix-up; Moonshot issued no public statement. This affects Kimi generally and was not reported as specific to Deep Research. Reports: OECD AI incident monitor, Phoenix Weekly (page unavailable as of 2026-10-17), Sina / Cover News, Seoul Economic Daily. No regulator action against Kimi was found.
Privacy note
Kimi is operated by Moonshot AI, a China-based company. Its privacy policy has not been reviewed for this entry, and its data handling and storage have not been independently verified.
Uncertainties
- Only two independent evaluations found; one is a preprint with competitor co-authors.
- Whether Deep Research now uses newer Kimi models (K2.5–K3) is not stated; the help page still names Kimi-Researcher.
- The résumé leak was reported by media; Moonshot issued no public statement.
- Data storage and privacy practices of the China-based service were not independently verified.
- Rechecked in Decision 0076 (19 Oct 2026): no new independent test of Kimi Deep Research found; DeepResearchEval, DeepResearch Bench II and FINDER do not test it, and the DeepResearch Bench leaderboard still shows no FACT score.
- The only citation-accuracy figure (20.80%, DualGraph) re-scores the DeepResearch Bench reports, whose citations the benchmark authors could not parse; it may not reflect citation quality.
Sources
Independent performance sources (counted)
- DeepResearch Bench (ICLR 2026; USTC): official UI, 29 Jun–1 Jul 2025; RACE 44.64 (comp. 44.96, depth 41.97, inst. 47.14, read. 45.59); no FACT scores (citations not parseable) — independent performance, accessed 2026-10-17
- ICBCBench (preprint 2026; ICBC, SJTU, Fudan, Shanghai AI Lab, Alibaba and others): Kimi-deep-research EN overall 47.59, ZH 44.72; 3rd of closed systems in English. Disclosure: Alibaba co-authors (Alibabas Qwen deep research is also evaluated); the paper does not state how or when Kimi was accessed — independent performance, accessed 2026-10-17
Other sources (not counted)
- OECD AI incident: Kimi disclosed a user's résumé to another user (20 April 2026); see also Phoenix Weekly http://www.ifengweekly.com/detil.php?id=26914 and Seoul Economic Daily — other (privacy incident, Kimi-wide), accessed 2026-10-17
- DualGraph, "A Tale of Two Graphs" (preprint, Feb 2026; Microsoft, Chinese Academy of Sciences, South China University of Technology): re-evaluated the DeepResearch Bench reports with a GPT-4.1 judge; Kimi (Res.) RACE 46.50, FACT citation accuracy 20.80%, 2.30 effective citations per task. Same reports as DeepResearch Bench, so not a new test; authors propose a competing method — other (re-scoring of DeepResearch Bench reports), accessed 2026-10-19
Not counted
- WebResearcher (Alibaba Tongyi): Kimi-Researcher rows taken from vendor reports — not counted (vendor numbers reused), accessed 2026-10-17
- Moonshot’s own benchmark results (Humanity’s Last Exam 26.9% pass@1, xbench-DeepSearch 69%, FRAMES, Seal-0, SimpleQA): vendor-reported, not independent.
Vendor statements
- Launch post (20 June 2025): HLE 26.9% pass@1, xbench-DeepSearch 69% (vendor-run; excluded) — vendor or maintainer, accessed 2026-10-17
- Help page: Deep Research powered by the Kimi-Researcher model; 10–25 minutes; cited report — vendor or maintainer, accessed 2026-10-17