Algoverse AI Research Program is a 12-week online research mentorship program placing students in teams of 3-4 with a dedicated mentor from institutions including Google DeepMind, Meta FAIR, OpenAI, Stanford, and CMU. Teams follow a three-phase process: identifying a research problem, conducting original research, and preparing a manuscript for submission to top AI venues including NeurIPS, ICML, ICLR, ACL, and EMNLP.
The program runs with two weekly meetings and an optional office hours session, requiring a commitment of 5-10 hours per week. GPU and compute costs are covered in the tuition. Past cohorts have achieved a 68-73% conference acceptance rate, with 311 students accepted to NeurIPS conferences and workshops. Student alumni have gone on to university admissions, fellowships, and internships, and Algoverse research has been cited by OpenAI, Microsoft, MIT, Princeton, and Oxford.
Multiple cohorts run throughout the year with staggered start dates, allowing students to choose a schedule that fits their availability.
Cycle and price check — November 4, 2026: the official team-program page publishes the Spring and Summer 2027 ranges retained above and a $4,325 team-program fee including compute. That replaces the old $3,325 catalog price. The 2026 prospectus also lists $4,325 for the 12-week team program and $8,325 for the separate 16-week 1:1 program.
The official page labels the November 1 cohort Winter 2026. The page continues to mark it Open although the start is past; this is not treated as confirmation that late enrollment is accepted. The old October 4 deadline could not be reverified and remains unverified catalog history. The page explicitly states rolling application review, but publishes no fixed deadlines for the listed 2027 sessions. Coming Soon cohorts are announced dates, not confirmed open application windows.
The official FAQ says Fundamental AI Research applicants need an introductory programming course and Applied AI Research has no programming prerequisite. This differs from the program page and prospectus's blanket Python-experience requirement; both are disclosed rather than imposing a single unsupported rule.