China Undergraduate Mathematical Contest in Modeling: Registration, Dates, and Rules

Overview

Divisions, problems, and participation

The contest has two divisions: undergraduate teams choose one of Problems A, B, and C, while vocational-college teams work on Problems D and E.

In 2025, Problem A covered smoke-screen projectile deployment, B concerned silicon-carbide epitaxial-layer thickness, C analyzed NIPT timing and fetal abnormality detection, D addressed mine water inrush, and E covered AI-assisted physical fitness assessment. The three undergraduate problems leaned toward physical modeling, engineering parameters, and data analysis respectively, so match the choice to your team's strengths.

Each of the last two contests drew more than 65,000 teams, and national first and second prizes together went to under 3% of undergraduate teams. The figures below come from official award announcements.

Timeline

Annual timeline: registration through awards

1. 01

   Around April: contest notice and university registration

   After the annual notice is published (the 2025 notice appeared on China Student Online in April), form a team of no more than three students from the same institution and register through the institution. Individuals cannot register directly; check your university's own deadline.

2. 02

   September: Thursday 18:00 to Sunday 20:00

   The contest lasts 74 hours. In 2025 it ran from September 4 at 18:00 to September 7 at 20:00; in 2026, from September 10 at 18:00 to September 13 at 20:00. Obtain the problems from the official channel named in that year's notice.

3. 03

   October: award list published

   After regional and national review, the award list is published. Verify your team and award information against the official announcement.

4. 04

   December: award ceremony and outstanding papers

   Awards are presented, and outstanding papers and problem-author commentaries are then released on the contest website and China Student Online. They are the most direct preparation material for the next contest.

Rules

Team and compliance requirements

These four rules summarize the contest notice and rules. Confirm each one with the whole team before the contest starts.

74 hours

Inside the 74 hours and what to submit

Official documents do not prescribe a team's internal schedule. The three-stage plan below makes progress easy to check; adjust it to the problem type and team roles.

1. 01

   D1 · Thursday evening: compare and select a problem

   Compare the data, workable methods, and team skills for each problem, then decide quickly. Switching later costs both modeling and writing time.

2. 02

   D2 · Friday to Saturday: build the model and get the core results

   Complete the main model, the solution, and the core numerical results. By Saturday evening you should have numbers ready for the paper, leaving day three for the main text, abstract, and validation.

3. 03

   D3 · Before Sunday 20:00: finish writing and check the format

   Add model validation and sensitivity analysis, make the abstract state the problem, method, main results, and conclusions, then check every format requirement before submitting.

• The electronic paper is a single PDF or Word file of no more than 20 MB.
• Supporting materials go in one RAR or ZIP archive of no more than 20 MB.
• Under the 2026 revision, the printed main text begins on page four, has no table of contents, and runs to no more than 30 pages.
• The appendix lists the supporting-material files and includes complete, runnable source code.
• The specification is revised every year; follow the current year's revision.

Judging

Judging: top one-third regionally, then national review

Judging has two levels. Regional review comes first, with awards generally going to no more than one-third of entries in each region. Outstanding regional papers then go to national review, which follows the national review specification across five stages and includes similarity checks.

Provincial awards therefore depend on the regional review, while national awards require passing national review. The two targets call for different levels of preparation.

Preparation

Practice past problems and set up your modeling environment with Scientify

Scientify is a science agent that runs in an isolated cloud computer. Give it a past problem and its attachments, and it reads the problem, builds the model, writes and runs solver code, recomputes the results, and writes the paper. The problem, code, dependencies, data, logs, and results stay in one workspace; the task keeps running after you close your laptop, and you can check progress from your phone.

On 2023 CUMCM Problem A (heliostat-field optimization), Scientify independently went from problem to paper across 131 real command executions. Its Problem 3 output per unit area was 0.459488 kW/m², above the 0.4556 kW/m² of official showcase paper A0165, reaching a national first-prize level. The full process is documented on the national contest case page.

• Have the agent work a past problem end to end, then compare it with official outstanding papers and problem-author commentaries to see where model choices differ.
• Get your usual solver, optimization, and plotting code running in the workspace before the contest so you spend less time on setup.
• Review the code and recomputation records the agent leaves, and make sure every teammate can explain the assumptions and where each result comes from.
• During the real contest, follow the current notice on whether and how AI may be used, and submit an accurate AI use report.