Mifeng Technology launched Mifeng Pai on September 23, 2026, according to 36Kr; China Securities Journal and Gasgoo independently reported the launch and corroborated selected product details. 36Kr reported the workflow and pilot disclosures, while the secondary reports supplied additional hardware, task, scenario, and internal-test details. Mifeng positioned the platform as a way to scale diverse environment-action data without deploying robots for every capture task, and it disclosed early internal-test metrics it said were achieved during a one-month pilot.
Readers can consult 36Kr’s report for Mifeng’s product narrative, workflow, and pilot figures, and China Securities Journal’s report for independently reported hardware and app-task examples.
The launch is described in 36Kr’s account of Mifeng Pai and China Securities Journal’s report. The current event is the September 23 public launch of a crowdsourced, robot-free collection operation.
How the platform works and early metrics
Mifeng described a capture workflow that starts with an app download, the rental or short-term use of an MEgo capture device, acceptance of a task, field execution, upload, quality review, and post-settlement remuneration based on validated data duration, according to 36Kr. China Securities Journal independently reported that the hardware package includes elements such as a two-finger gripper, a head-mounted vision unit, and a wrist camera, which together comprise the robot-free capture approach the company calls Ego data. Pre-launch coverage in Securities Times noted the September 23 public release date ahead of the launch, and 36Kr published details after Mifeng’s announcement.
Mifeng told 36Kr that after one month of internal testing the platform had 20,000 registered users and 13,000 task submissions, and that the top internal tester earned more than 5,000 yuan in a month. Those figures are company disclosures from the internal pilot and should not be taken as audited usage or revenue metrics, since registration does not equate to sustained activity. Gasgoo reported more than 20,000 registered users and 13,000 submissions after one month of internal testing; its headline described this as users participating, but the opened sources do not establish active participation. Gasgoo’s scenario taxonomy overlaps with, but is not identical to, 36Kr’s account.
Task types, pay mechanics, and quality controls
Mifeng described task windows spanning brief captures and sessions lasting longer than sixty minutes, and said the platform may halt distribution once a data type is fully collected. 36Kr reported platform pricing shown in the app, including an MEgo rental fee of 39 yuan per day and a promotional rate of 19 yuan per day, and a displayed basic return in the neighborhood of 20 yuan per valid hour. Mifeng framed compensation as a mix of baseline urban pricing and dynamic subsidies tied to efficiency and data quality, with final settlement determined by verified valid duration; these are company-reported platform terms and not guaranteed wages.
China Securities Journal offered concrete app examples illustrating how task length maps to expected pay. One logistics loading and handover capture was observed with a 2 to 90 minute range and an estimated return of 0.67 to 30 yuan, while a stranded-parcel cleanup example showed a 2 to 20 minute range with an estimated 0.67 to 6.67 yuan return. These samples were reported as app-displayed estimates and should not be read as a universal rate card for all tasks or locations.
On data processing, 36Kr outlined Mifeng’s screening pipeline: submissions are segmented into actions, labeled, and processed for trajectory extraction and manual sampling inspection, after which surviving data are graded by quality and use value. Mifeng’s chairman and CEO, Yao Maoqing, told 36Kr that more than 95 percent of data passing initial settlement screening remains usable after post-processing; that proportion is a company claim and not an independent audit. The company said it classifies data not only as usable or unusable but by downstream roles including pre-training, specific-task training, evaluation, and long-tail scenario coverage.
Market rationale, scaling plans, and uncertainties
Yao Maoqing told 36Kr that a single customer’s training demand can reach millions of hours, while some requests extend to tens of millions. He said leading customers can expect suppliers to provide between 50,000 and 100,000 hours each week. These are executive statements about reported demand, not independently verified market totals. These figures were presented as company observations of customer requests and should be treated as executive statements about demand rather than independently verified market totals. Mifeng framed its role as a turnkey data-service provider that can handle equipment, manpower, scenarios, and data operations so customers can concentrate on hardware and model research, according to 36Kr.
To address scenario access, Mifeng said it plans a Scenario Data Alliance intended to link factories, communities, hotels, stores, and logistics centers, and the company reported that the initiative was co-initiated with more than 50 enterprises and institutions. Gasgoo also reported a first group of over 50 organizations across hotels, catering, supermarkets, elder care, medical care, manufacturing, and urban property. Mifeng said it will conduct regular production-plan review meetings to map market orders against inventory and to route missing data to different collection channels based on order value and available resources. These are planned or stated operational practices and not confirmations of completed long-term partnerships.
Mifeng acknowledged capital and infrastructure scale are material constraints. The company told 36Kr that supporting tens of millions of hours would require hundreds of millions of yuan in fixed assets for devices, personnel settlements, storage, and compute. Yao also suggested that data-supplier concentration could increase as customers prioritize supplier scale, quality, and delivery; that forecast is an executive opinion, not a market conclusion.
The platform launch leaves open several uncertainties that matter for prospective participants and customers. It is not yet independently verified how representative the internal-test metrics are of long-term participation or how device rental and task returns translate into net participant income after expenses. Quality claims come from company disclosures and have not been publicly audited. The Scenario Data Alliance membership and the mechanics of data authorization and privacy compliance remain at the planning and early-commitment stage as reported by Mifeng and corroborated in secondary coverage.
Mifeng Pai introduces a public-participation route to gather embodied-intelligence training data at scale, but key questions about sustainable participant economics, independent validation of data quality, and the practicalities of large-scale scenario access remain unresolved. The platform’s initial disclosures describe a structured capture and review pipeline, promotional pricing, and an early pilot with thousands of registrations; whether those early signals will scale into reliable, high-volume delivery for demanding customers will depend on execution, verification, and the degree of industry cooperation Mifeng can secure.