// side by side · same rules, same day

Operator vs Manus

Operator 4/5 · Manus 5/5

Download the card Operator dossier →Manus dossier →

The three axes

	OperatorOpenAI AGENT	ManusMonica AGENT
CAPABILITYyes / no / uncertain	4/5 AGENT	5/5 AGENT
PROOFhow well you can check us	OPERATOR INSPECTABLE	OPERATOR INSPECTABLE
TRACE ACCESSwho can read the record	OPERATOR-OWNEDWhoever runs it holds the record; nothing is published	OPERATOR-OWNEDWhoever runs it holds the record; nothing is published
LEASHtool calls between human turns	not measured — measure it yourself	not measured — measure it yourself
CHECKEDversion and date	Operator research preview (Jan 2025); folded into ChatGPT agent, standalone site sunset July 20252026-08-11	Manus (Cloud Computer / Browser Operator era docs, Aug 2026)2026-08-11
CONFIDENCEours, in this dossier	4/10	8/10
COMMUNITYwhat visitors ruled	A0 · W0 · C0n=0 raw votes · signal at 25	A0 · W0 · C0n=0 raw votes · signal at 25
IN COURTopen challenges	no cases filed	no cases filed
DISPUTEDwhere the axes disagree	nothing disputed — every answer is inspectable	nothing disputed — every answer is inspectable

Question by question

rows marked "they differ" are where the comparison earns its keep

	OperatorOpenAI AGENT	ManusMonica AGENT
LOOPSruns until done	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 CUA repeats a screenshot→chain-of-thought→click/type loop "until a given task is completed or further input is required"; the 400-step eval cap is a ceiling, not a fixed script.	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 Vendor's context-engineering post: a typical task averages ~50 tool calls, observations fed back in.
CHOOSESpicks its own tools	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 OpenAI states the model "can choose to open a page using the text browser or visual browser," then run a terminal command — tool and next UI action picked at runtime from the current screen.	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 Blog: rather than a preset order, Manus masks logits at decode so the model picks each next tool.
ACTSchanges the world	✓ yes OPERATOR INSPECTABLE confidence 9/10 · checked 2026-08-11 Drives a real browser with mouse/keyboard on live sites — clicks, fills and submits forms, orders groceries and tickets — plus terminal commands and file writes on its virtual computer.	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 Docs: a sandboxed virtual computer with persistent files; it installs software and drives a browser.
RECOVERSfixes its own errors	✓ yes OPERATOR INSPECTABLE confidence 7/10 · checked 2026-08-11 System card p.20: after failing to install a biodesign tool, the agent "researched and wrote substitute scripts" and continued unaided — a different route it chose, not builder-wired retry.	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 Blog: failed actions and error traces stay in context to adapt; it calls error recovery core to agency.
UNSUPERVISEDnobody watchingthey differ	✗ no OPERATOR INSPECTABLE confidence 4/10 · checked 2026-08-11 Watch mode pauses when you go inactive, and it asks confirmation before purchases or sending mail.	✓ yes OPERATOR INSPECTABLE confidence 8/10 · checked 2026-08-11 Docs: scheduled tasks run automatically with no approval gate; it only prompts for logins/CAPTCHAs.

Every answer here is the same one on the dossier and in the dataset — this page holds no numbers of its own. Think one is wrong? Bring a trace.