HIVE RESEARCH · IMO 2026 · HUMAN × AI

Six problems. One verifiable Human–AI journey.

An open record of how a human and AI read, attempt, fail, repair and verify all six IMO 2026 problems.

The IMO does not only test finding an answer. It tests seeing structure, choosing the right lemma, and writing a proof with no hidden steps. HIVE-IMO X records the whole process - blind reading, wrong routes, counterexamples and proof candidates - through to independent verification.

This is not an “AI solved mathematics” showcase. It is a record of how human judgment, machine search, and verification work without collapsing into one another.

Live dashboard

Six problems, two exam days

IMO 2026 is the 67th IMO, held in Shanghai, 10–21 July 2026. Exam days are 15 and 16 July, three problems each. A problem counts as complete only at VERIFIED-INDEPENDENT or beyond.

6/6
Legacy solution manuscripts
written earlier; not independently verified
0/6
Official statements fully reviewed
S0 source gate, pending V2/V3
0/6
Independently audited proofs
the only figure that counts as complete

These three figures are different. Six complete solution manuscripts already exist from earlier work; none has yet passed independent audit under this public pipeline.

Day 1 · 15 Jul 2026 · P1 · P2 · P3
Day 2 · 16 Jul 2026 · P4 · P5 · P6
0/6
Independently audited proofs
the only 'complete' figure
0
Proof candidates rejected
0
Counterexamples found
0
Lemmas retained
Not yet logged
Human time
Not yet logged
AI compute time
The method

Every problem passes through fifteen stages

The pipeline separates reading, attempting, verifying and comparing so no single agent can propose and bless its own answer.

S0
Official problem freeze
S1
Human blind read
S2
AI independent read
S3
Contrastive merge
S4
Formal decomposition
S5
Small-domain reconnaissance
S6
Route tournament
S7
Lemma forge
S8
Candidate proof assembly
S9
Adversarial review
S10
Independent reconstruction
S11
Human proof rewrite
S12
Independent / formal verification
S13
Official-solution comparison
S14
Publication and postmortem
Operating model

Three roles, kept separate on purpose

Lâm Nguyễn
Human Owner / PI
  • Choose research direction
  • Hold intuition and the natural standard of proof
  • Approve gates
  • Control claims
  • Own publication
Sol
AI Contractor · ChatGPT 5.6 Thinking
  • Formalize problems
  • Map the theorem landscape
  • Design gates
  • Red-team proofs
  • Surface hidden assumptions
  • Issue verdicts
Fable
AI Builder · Claude Code
  • Implement search / verifier
  • Run tests and counterexample search
  • Produce artifacts
  • Update the website from the approved ledger
No single agent may propose, implement, verify, and declare success for the same claim.
The 6/6 summary is locked

It unlocks only when all six problems reach at least VERIFIED-INDEPENDENT. The project will not display “6/6 solved” before then.

0/6

HIVE-IMO X is an independent research project. Not affiliated with or sponsored by the International Mathematical Olympiad. Problem texts belong to their authors. Human Owner Lâm Nguyễn holds academic responsibility; Sol and Fable are AI collaborators, not accountable authors.