Manus
- Category
- Research
- Type
- commercial
- Scope
- General-purpose agent (research and multi-step tasks)
- Source
- Vendor site
- Last reviewed
- 2026-10-18 (what this means)
Summary
Manus is a commercial general-purpose AI agent used for research reports and multi-step digital tasks. Meta agreed to acquire it in December 2025; Chinese regulators ordered the deal unwound in 2026, some user data was deleted in August 2026 to comply with regulatory requirements, and Manus resumed independent operation in September 2026. Manus 2.0 (28 September 2026, new architecture) has not been independently tested, so no overall score is given. Independent tests of 2025 versions are listed as historical context only. 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: 0 for Manus 2.0. Manus 2.0 (28 Sep 2026, new architecture) has not been independently tested.
Factor notes (no scores)
| Factor | Weight | Score | Note |
|---|---|---|---|
| Completion rate | 30% | — | No independent test of Manus 2.0 was found. 2025 versions, for reference only (not counted): research-report quality mid-range with weak citation accuracy (LiveResearchBench citation 45.6; DeepResearchEval quality 5.95/10, 8th of 9, factual ratio 82.3%, 1st of 9); 8.0% full success on CAP cross-site tasks (best agent tested; humans 10%); fabricated all 10 selected studies in a systematic-review test; 92–95% accuracy on dental diagnosis questions and radiographs. |
| Disputes | 20% | — | Researchers demonstrated prompt injection exposing a VS Code server (Embrace The Red, 2025) and a single email leading to remote code execution and exposure of connected-account tokens (Salt Labs, 2026; fixed via Meta's bug bounty). Chinese regulators ordered Meta's acquisition unwound, and some user data was deleted in August 2026 to comply with regulatory requirements. |
| Adherence to limits | 25% | — | In both security demonstrations, Manus followed instructions hidden in content it processed. Not independently tested for Manus 2.0. |
| Evidence quality | 25% | — | Zero independent performance sources for Manus 2.0. Six tests of 2025 versions (five peer-reviewed) are listed but not counted. |
See the methodology for the insufficient-evidence rule.
Ownership and product timeline
- 15 December 2025: Manus 1.6 released.
- 29 December 2025: Meta agreed to acquire Manus. In 2026 Chinese regulators ordered the deal unwound.
- 23–24 August 2026: some user data generated between 29 December 2025 and 23 August 2026 was deleted “to comply with regulatory requirements”, per the service change notice.
- 1 September 2026: Manus resumed independent operations.
- 28 September 2026: Manus 2.0 released; the vendor says it “is not a version update. It’s a new architecture.”
Scope
This entry covers Manus as currently offered (Manus 2.0). Independent tests of 2025 versions are listed below as historical context only and are not counted, because the vendor describes Manus 2.0 as a new architecture rather than a version update.
Uncertainties
- All independent tests found used 2025 versions; Manus 2.0 is a new architecture per the vendor and is untested.
- Vendor-reported GAIA scores and Manus 2.0 efficiency figures are not used.
- The two dental studies test diagnostic question answering rather than agentic tasks.
- CAP has a competitor tie (one author interned at FellouAI, an evaluated system).
- Manus 2.0 Cascade efficiency figures (23.2% fewer tokens, 28.2% less time, 32% lower cost; one vendor configuration) are not used.
- Cue agents (own email, phone number, wallet; invite-only early access) have not been independently tested.
Sources
Historical context: tests of 2025 versions (not counted)
- LiveResearchBench (ICLR 2026; Salesforce, UW-Madison, Stanford): presentation 75.0, fact and logic 63.1, coverage 73.3, citation 45.6; LLM judges; 2025 — historical context (earlier version, not counted), accessed 2026-10-17
- DeepResearchEval (preprint 2026): 100 tasks; quality 5.95/10, 8th of 9; factual ratio 82.30%, 1st of 9; reports collected 2025 — historical context (earlier version, not counted), accessed 2026-10-17
- CAP (COLM 2026): 23.0% partial completion, 8.0% full success, best agent tested (humans 10%); Nov–Dec 2025. Competitor disclosure: one author interned at FellouAI, whose system was also evaluated — historical context (earlier version, not counted), accessed 2026-10-17
- Loke et al. (Eye 2026): April 2025 systematic review and meta-analysis test; only agent attempting the full workflow, but all 10 selected studies and 5 references were fabricated — historical context (earlier version, not counted), accessed 2026-10-17
- Madfa et al. (Frontiers in Oral Health 2025): March 2025 version; 92.3% on 117 dental case questions (differences from other models not significant) — historical context (earlier version, not counted), accessed 2026-10-17
- Annals of Medicine 2026: April 2025 version; 92.5% then 95.0% on 120 dental radiographs (radiologists 96.7%) — historical context (earlier version, not counted), accessed 2026-10-17
Other sources (not counted)
- New Runtime analysis (October 2026): Manus 2.0 Cascade efficiency figures are vendor-reported for one configuration; no independent benchmark — other (commentary, no original testing), accessed 2026-10-18
- MIT Technology Review hands-on review, March 2025 (qualitative) — other (hands-on journalism, earlier version), accessed 2026-10-17
- Embrace The Red (2025): prompt injection led Manus to expose its VS Code server to the internet; reported 1 June 2025 — other (security test of earlier version), accessed 2026-10-17
- Salt Labs (published 1 October 2026; research conducted earlier in 2026, before Manus 2.0): a single email led to remote code execution and exposure of connected-account tokens; fixed via Meta bug bounty — other (security test), accessed 2026-10-17
Vendor statements
- Manus 2.0 announcement (28 September 2026): "not a version update. It's a new architecture" — vendor or maintainer, accessed 2026-10-17
- Service change notice: some data generated 29 Dec 2025–23 Aug 2026 deleted 23–24 Aug 2026 to comply with regulatory requirements — vendor or maintainer, accessed 2026-10-17
- Manus resumes independent operations (1 September 2026) — vendor or maintainer, accessed 2026-10-17
Excluded sources
- Manus GAIA scores (86.5 / 70.1 / 57.7) and Manus 2.0 efficiency figures: vendor-reported, not independent.