The Cycle Double Cover Conjecture has been open since 1966. GPT-5.6 Sol Ultra, released July 9, produced a formal proof by July 10. The PDF is live on OpenAI's CDN, the Hacker News post scored 487 points, and the academic community has not yet issued formal verification — but the speed and specificity are worth noting.

This is not a benchmark game. This is a named problem with a sixty-year academic pedigree, solved by a model in general release. The IMO 2026 market on Manifold is now at 88¢ YES for a perfect AI score, up from mid-sixties two days prior. The 24-hour volume is $12,567, which is real money for a Manifold contract.

For institutional readers: if your technical hiring plan assumes human mathematicians retain a meaningful edge on novel proof construction, that assumption is now live-fire tested. The model is not vaporware. It is in production, it is solving problems that tenured faculty could not, and it is doing so on a public release cycle. Comp bands for ML research roles and proof-adjacent engineering should be revisited before Q3 planning locks.

The complaint was filed Friday in the Northern District of California. Apple names three former employees now at OpenAI and alleges misappropriation of proprietary hardware development processes. The timing is notable: two days post-GPT-5.6 release, and concurrent with $257,134 in 24-hour Polymarket volume across OpenAI consumer hardware contracts.

The hardware markets are live but unreadable — the yes/no price fields in the data feed are returning bracket characters instead of decimals, which suggests either a data formatting issue or a market suspension. What is clear: three contracts covering 2025, March 2026, and December 2026 launch windows are all actively traded, and the volume is institutional-scale.

Apple does not file unless the exposure is real. OpenAI does not staff a hardware division unless the product is real. The prediction markets are pricing non-zero probability across multiple tenors, and the lawsuit adds discovery risk to any forward launch timeline. For compliance and procurement teams: if you are in active RFP discussions with OpenAI on hardware or embedded AI offerings, the litigation calendar is now part of your delivery risk model.

The GPT-6 release market for July 31, 2026 is currently priced at 91¢ NO, meaning the market assigns a 9% probability to a launch in the next twenty days. OpenAI released GPT-5.6 on July 9. The company has never shipped two integer model versions within a single month, and there has been no public guidance suggesting a July GPT-6 date.

The market is pricing correctly. The volume — $50,939 in 24 hours — suggests participants are using this as a high-conviction short, which it is. The August 7 contract is also at 91¢ NO, and the August 31 tenor shows similar structure. The curve is flat because the base case is clear: GPT-6 is not imminent.

But the March 2026 contract is missing volume data in the feed, and the December 2025 contract is also blank. That is where the asymmetry lives — not in the twenty-day window, but in the six-to-twelve month structure where OpenAI's actual release cadence would typically land. For readers building H2 model deployment roadmaps: the July/August contracts are consensus short, the December and Q1 2026 tenors are where the real positioning will emerge, and right now we do not have the tape.

The data is broken or the markets are suspended. Polymarket's three OpenAI hardware contracts — 2025, March 2026, December 2026 — show $41,235, $184,752, and $31,146 in 24-hour volume respectively. The yes_price and no_price fields are returning "[" and '"' instead of decimal probabilities. That is either a feed formatting error or a market halt.

What we know: six-figure daily flow is real money, the Apple lawsuit filed Friday cites hardware timelines, and OpenAI has never publicly announced a consumer device. The markets exist because the rumor has institutional distribution. The volume exists because the rumor has institutional belief. The pricing is unreadable because something in the data pipeline is malformed.

For procurement and strategic planning teams: the prediction markets are treating OpenAI hardware as a live question with near-term resolution probability, but we cannot currently read the spread. If you are in RFP discussions with OpenAI, if you are planning enterprise hardware refresh cycles that assume Apple-only or Google-only AI endpoints, or if you are modeling consumer AI device market share for 2026-2027, the market is telling you the question is live — it is just not telling you the price.

The International Math Olympiad 2026 takes place in mid-August. Manifold's market on a perfect AI score is now at 88¢ YES, up from approximately 69¢ before GPT-5.6's release. The 24-hour volume is $12,567, which is significant for Manifold's liquidity profile. The proof of the Cycle Double Cover Conjecture is the catalyst — if a model can close a sixty-year open problem in seventy-two hours, a timed Olympiad problem set is well within range.

The market is pricing correctly, but it is also pricing five weeks of model improvement, potential access restrictions, and contest rule interpretation. The IMO has not yet published official guidance on AI participation or scoring, and the prediction market is assuming the question resolves based on technical capability, not regulatory permission.

For institutional readers: this is a benchmark market with a hard deadline and a clear resolution mechanism. If you are allocating to AI model development, if you are underwriting technical talent with Olympiad credentials, or if you are modeling the delta between human and model performance on high-stakes reasoning tasks, the market is telling you the delta has collapsed. The contest is in five weeks. The proof was published two days ago. The probability curve is steep and it is still moving.