AI Perfect Score at IMO 2026? Manifold Traders Sit at 94% as Shanghai Grading Runs. A Manifold market asking whether an AI model will post a flawless 42/42 at the 2026 International Mathematical Olympiad is trading at a 94% implied probability of YES as of July 19, 2026, with 513 unique bettors and top-tier 24-hour volume on the platform. The move follows this year’s exam papers in Shanghai on July 15 and 16 and a run of 2025 results that took machines from silver to gold in a single season. Within days, Huawei said its Celia model and Xiaohongshu/RedNote said its dots-note 3.0 model had each turned in a 42/42 run, the first reported perfect AI scores in the olympiad’s history, though neither claim yet carries official IMO certification, which is exactly why the money is interesting.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the Manifold market is actually asking
The contract in question is “Perfect score achieved by an AI model in the International Math Olympiad (IMO) 2026?”, created by the user Bayesian on Manifold. A perfect score at the IMO means 42 out of 42 points: six proof problems worth seven points each, split across two exam days of roughly 4.5 hours apiece. To resolve YES, the market looks for an AI model to fully solve all six 2026 problems under conditions close to the human contest, with published and checkable results rather than a press-release claim.
That framing matters because it sets a bar well above what any system has publicly cleared. Reaching gold in 2025 required 35 points and five solved problems. A perfect score requires the sixth, which is the problem that stopped nearly every human in the room and every reported machine last year. If you are new to how these contracts price outcomes, our primer on how prediction markets work walks through the mechanics.
The odds as of July 19, 2026, and what they imply
As of July 19, 2026, the Manifold market prices YES at 94%, with 513 unique bettors holding positions. On a platform that runs on play-money mana rather than cash, the number is best read as a crowd forecast rather than a wager on real dollars. A 94% reading translates to roughly 15-to-1 odds that a model lands the perfect run, a level of confidence that would have looked absurd two years ago.
Implied probability is simply the market price restated as a percentage chance, and it is not a guarantee. Our explainer on implied probability covers why a 94% line still leaves real room for a NO outcome, and why late-stage markets on binary events can swing hard when a single data point lands. This particular market has climbed through the summer, and Tech Insider tracked earlier prints when the same contract sat near 96% and, before that, in the low 90s.
Why the market is moving now
The timing is not random. The 67th International Mathematical Olympiad is being held in Shanghai, People’s Republic of China, with the competition window running July 10 to 21, 2026, and the two exam papers held on July 15 and 16, per the official IMO 2026 edition page. Grading by contest coordinators takes place after the papers, so the days right after the exams are when speculation peaks: the problems are known, the human scores are being finalized, and the AI labs that ran the same papers have a window to publish.
Traders were betting into that gap. With the 2026 problems public and 666 human participants sitting the same papers in Shanghai, anyone could watch how the strongest models handled them, and the surge in bettors and volume tracked the post-exam news cycle rather than any confirmed result. By late July 2026, Huawei’s Celia and Xiaohongshu/RedNote’s dots-note 3.0 had each been reported to hit 42/42, but as of July 19, 2026, no lab had published a 2026 perfect score certified by official IMO coordinators.
The 2025 breakthrough that set the stage
The confidence baked into this market traces back to July 2025. At the IMO in Queensland, Australia, two systems reached the gold threshold of 35 out of 42. Google DeepMind reported that an advanced version of Gemini with Deep Think solved five of six problems and had its work graded and certified by IMO coordinators using the same criteria applied to students. OpenAI separately said an experimental reasoning model hit the same 35-point mark, though its solutions were graded by three former IMO medalists rather than official coordinators.
Context keeps the milestone honest. As CBS News reported, 72 human contestants earned gold in 2025, 45 of them scored exactly 35 like the machines, and 26 scored better than either program. Both systems also stalled on the same sixth problem. Humans still won the room, but the gap narrowed fast, and coverage from outlets including Gizmodo framed 2025 as the year AI reached the podium.
The open-source angle: DeepSeek Math-V2
The story is no longer confined to two US labs. DeepSeek released Math-V2, an open-weight model that reached gold-level performance on the 2025 IMO problems by solving five of six, making it the first openly available system to do so, according to the South China Morning Post. The model was published on Hugging Face and GitHub under a permissive license, and it uses a generator-verifier loop that checks its own proofs rather than only producing an answer.
For a finance reader, the open-weight release changes the shape of the bet, and the compute economics sharpened the picture further. Reports in July 2026 put GPT-5.6 Sol xhigh at a perfect score in 3.8 hours for roughly $20, Claude Fable 5 at 2.5 hours for $51, and K3 at 17.4 hours for $31, meaning gold-level math reasoning could now be rented for double-digit dollars rather than a lab’s entire compute budget. When gold-level math reasoning can be downloaded, run by third parties, or bought outright for the price of a dinner, the number of teams that can throw compute at the 2026 problems grows, and so does the chance that someone publishes a verifiable full solution. That broadening of the field is part of why the crowd sits so high.
The benchmark evidence traders are watching
Beyond the IMO itself, the closest read on model strength comes from MathArena, an independent site that grades models on fresh, uncontaminated competition problems. On the 2026 USA Mathematical Olympiad, a proof-based contest one tier below the IMO, GPT-5.4 scored about 95%, with Gemini 3.1 Pro next at roughly 74%. A year earlier, models had performed disastrously on the 2025 USAMO, producing circular arguments and unsupported guesses. By August 2026, a benchmark snapshot had gone a step further, listing Claude Opus 5 with a full 42/42 score on the IMO 2026 problems themselves. The jump from near-zero to near-saturation in thirteen months is the single strongest argument the bulls have.
Saturating the USAMO is not the same as a perfect IMO, and MathArena’s own graders have flagged that the hardest olympiad problems remain where automated proofs break down. But for a market pricing momentum, a near-95% USAMO print is the kind of signal that pulls the line toward YES.
The catch: verification and the IMO’s stance
The gap between a claim and a confirmed result is the whole ballgame here. In 2025, only DeepMind’s run was graded by official IMO coordinators; OpenAI’s was scored by outside medalists. That distinction fed a broader debate, covered by Scientific American, over how to validate lab claims when compute budgets, tool access, and reproducibility are not always disclosed. The IMO does not recognize AI systems as official participants, so even a genuine 42/42 would be unofficial unless the rules change.
That is the risk the NO side is buying. A market can price a perfect score at 94%, but resolution depends on someone producing a full, checkable set of six correct 2026 solutions under contest-like conditions, and on the market’s operator agreeing the evidence qualifies. Ambiguous grading, partial disclosure, or a stumble on the sixth problem could all keep the print from landing.
Timeline: how AI climbed the olympiad ladder
| Date | Milestone | Result |
|---|---|---|
| July 2025 | DeepMind Gemini with Deep Think, graded by IMO coordinators | 35/42, gold level |
| July 2025 | OpenAI experimental reasoning model, graded by former medalists | 35/42, gold level |
| Late 2025 | DeepSeek Math-V2, first open-weight model at gold level | 5 of 6 solved |
| 2026 | GPT-5.4 on USAMO 2026 (MathArena) | ~95% |
| July 15-16, 2026 | IMO 2026 exam papers, Shanghai | Grading pending |
| July 19, 2026 | Manifold “perfect score” market | 94% YES, 513 bettors |
What a tech and finance reader should watch next
Three things will decide this market. First, whether any lab publishes 2026 solutions with independent grading in the days after the exams, ideally by IMO coordinators rather than internal reviewers. Second, how the sixth problem behaves, since it has been the historical wall for both machines and most humans. Third, how the market’s operator interprets its own resolution criteria if a claim lands in a gray zone. Related crowd forecasts, such as our coverage of how many human contestants post perfect scores, give a useful cross-read on how hard the 2026 paper turned out to be.
For readers weighing where these contracts trade, the venue matters as much as the odds. Manifold runs on play money, while regulated and cash-settled markets differ sharply in access and treatment, a contrast we cover in our Kalshi vs Polymarket comparison.
Frequently asked questions
Has an AI actually scored a perfect 42/42 at IMO 2026?
No. As of July 19, 2026, no AI perfect score has been officially confirmed. The IMO 2026 exams were held July 15 and 16 in Shanghai, and grading was still in progress.
What does the 94% Manifold number mean?
It is the market-implied probability that traders assign to a YES outcome, as of July 19, 2026. It is a forecast on a play-money platform, not a confirmed result or investment advice.
What did AI achieve at the IMO in 2025?
Google DeepMind and OpenAI each reported gold-level scores of 35 out of 42, solving five of six problems. DeepMind’s run was certified by IMO coordinators; OpenAI’s was graded by former medalists.
Why is a perfect score harder than gold?
Gold required 35 points and five solved problems in 2025. A perfect 42/42 requires the sixth problem, which stopped nearly every human and every reported machine last year.
Would an AI perfect score be official?
No. The IMO does not recognize AI systems as official participants, so any AI result, however strong, would be unofficial unless the rules change.
The Bottom Line
Sources
- Manifold — Perfect score achieved by an AI model in the International Math Olympiad (IMO) 2026?
- IMO Official — International Mathematical Olympiad 2026 (Shanghai)
- Google DeepMind — Gemini with Deep Think achieves gold-medal standard at the IMO
- CBS News — Humans triumph over AI at annual math Olympiad, but machines are catching up
- South China Morning Post — DeepSeek releases first open AI model with gold-level scores at maths olympiad
- MathArena — USAMO 2026 leaderboard
- Scientific American — Mathematicians question AI performance at the International Math Olympiad
- Gizmodo — Google and OpenAI chatbots claim gold at International Math Olympiad
Prediction markets carry risk and are not investment or betting advice. Market prices are forecasts, not facts, and can move sharply or resolve against the crowd. Availability is restricted by jurisdiction: Polymarket is not available to US persons, and Kalshi is a CFTC-regulated US exchange. Manifold operates on play-money mana. Age limits of 18+ or 21+ apply where relevant. If gambling is a problem for you or someone you know, call 1-800-GAMBLER for confidential help.
