A Manifold market now prices a non-MF group proof at 70% after Astra’s math shock. A play-money question on whether mathematicians will prove the existence of a “non-MF group” this month jumped to a 70% implied chance of YES on Manifold as of August 18, 2026, with 32% of the market’s all-time volume changing hands in a single day. The trigger is not a new proof. It is the ripple effect of an OpenAI model that, two weeks earlier, cracked a closely related 27-year-old problem in group theory.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
The market at a glance
The contract lives on Manifold, a forecasting platform where users trade on outcomes using “mana,” the site’s play-money currency rather than cash. The question, posted by user SCS, is blunt: “Will the existence of a non-MF group be proved this month?” As of August 18, 2026, the market sits at roughly a 70% implied probability of YES, and Manifold’s activity log shows that about 32% of the contract’s entire lifetime volume traded in the prior 24 hours, placing it among the most active markets on the platform for the day.
Because Manifold runs on mana, no dollars are at stake here in the way they would be on a regulated exchange. That distinction matters for how you read the number. The 70% reflects trader conviction and attention, not capital at risk, and it can swing hard on thin volume. Compare that to how regulated financial markets get measured: the IAIS’s mid-year Global Insurance Market Report, released in July 2026 using 2025 data, described solvency, liquidity, and profitability across the insurance sector as broadly stable, the kind of grounded, audited assessment that a play-money contract like this one simply does not have behind it. Still, a sudden surge to a third of all-time volume in one day is a signal that something in the real world moved.
What a “non-MF group” actually is
The jargon needs unpacking. In abstract algebra, a group is a set with an operation that satisfies a few rules, and mathematicians study which infinite groups can be “approximated” by finite matrices. A group is called MF, short for matricial field, if it can be approximated by finite-dimensional unitary matrices in operator norm. Put loosely, an MF group behaves, up to tiny error, like something you could model on a finite computer. Worth flagging for anyone searching the term cold: “non-MF” is not exclusive to group theory. A BSE corporate filing from October 2025, for instance, reported “non-MF revenue” growing 4.4% year-over-year to reach 13.0% of total revenue, using MF there as shorthand for mutual funds, a completely unrelated bit of finance jargon that happens to share the same abbreviation.
Here is the catch that makes the market interesting. As reflected in standard references on the topic, no example of a non-MF group is currently known, and it is an open question whether every group is MF. Mathematicians have kept expanding the list of groups proven to be MF, including every amenable group, without ever exhibiting one that fails the property. So the market is not betting on a result that already exists. It is betting on whether someone produces the first counterexample, a group that provably cannot be approximated this way, within the calendar month.
The odds and what 70% implies
A 70% implied probability sounds confident, but the framing deserves scrutiny. The question asks for a proof “this month,” a tight window for a problem that has resisted a definitive counterexample for years. Read literally, the market is claiming that traders think it is more likely than not that a first-ever non-MF construction lands within weeks.
That is an aggressive prior, and it is worth separating two things a forecast can measure: the long-run chance that non-MF groups exist at all, and the near-term chance that a proof is published on a specific deadline. Those are very different bets. Prediction markets routinely blur them when a question is written loosely, and thin play-money markets are especially prone to momentum, where a headline pushes a handful of traders and the price overshoots. For context on how these implied numbers are built and where they mislead, see our explainer on reading prediction-market odds and our look at AI-benchmark markets.
What is driving the sudden bet
The catalyst is an artificial intelligence result, not a mathematics preprint about MF groups specifically. On August 1 and 2, 2026, OpenAI disclosed that an unreleased model it calls Astra had produced machine-checkable proofs for ten long-standing open problems across group theory, geometry, and complexity theory. The headline result, reported by TechTimes and SiliconANGLE, was the construction of the first known non-sofic group.
Soficity is a sibling concept to the MF property. A group is sofic if its multiplication can be approximated arbitrarily well by finite permutation systems. The question of whether a non-sofic group exists had been open since Mikhail Gromov introduced the notion in 1999. Astra’s answer was that such a group does exist, settling the question in the negative for the “all groups are sofic” hypothesis. The logic driving the Manifold traders is a direct extrapolation: if an AI system can build the first non-sofic group after 27 years, perhaps the analogous MF question, another “maybe every group has this property” problem, is the next domino. That is speculation about a tool, not evidence of a specific non-MF construction in the pipeline.
Who and what is involved
The named players are concrete. OpenAI published three documents to support the announcement, including a 249-page technical manuscript and proof certificates in Lean 4, a proof assistant that mechanically checks whether each logical step holds. According to reporting from The Next Web, the repository’s “sorry” count, Lean’s marker for an unproven gap, stood at zero, meaning every step across all ten proofs was formally verified. OpenAI’s head of mathematics research, Sebastien Bubeck, confirmed the results publicly, and the company put the total compute cost for the ten solutions at roughly $2,000 at its stated API rates, a figure highlighted by Forbes.
Beyond the non-sofic group, Astra’s list reportedly included disproving Connes’s rigidity conjecture and proving Ehrhart’s volume conjecture. None of the ten publicly reported results is a non-MF group. That absence is the single most important fact for anyone weighing the Manifold contract. The market is pricing a hypothetical that no named source has yet reported as done or imminent.
MF, sofic, and hyperlinear: a quick comparison
The three properties often get lumped together because they all ask whether an infinite group can be approximated by finite structures. They are technically distinct, and their open-question status differs after August 2026.
| Property | Approximated by | Counterexample known? | Recent status |
|---|---|---|---|
| Sofic | Finite permutations | Yes, as of Aug 2026 | First non-sofic group reported via Astra, Aug 1-2, 2026 |
| MF (matricial field) | Finite unitary matrices in operator norm | No | Open; every amenable group known to be MF |
| Hyperlinear | Finite matrices in normalized trace | No | Open; no counterexample reported |
The table shows why the bet is a leap rather than a certainty. Sofic fell first, but MF and hyperlinear remain open, and progress on one does not automatically transfer to the others.
Why a tech and finance reader should care
Two threads make this more than a niche math curiosity. The first is the AI-for-science story. A machine producing formally verified proofs of decades-old problems for about $2,000 in compute is a data point about where automated reasoning is heading, and it is the same trend investors are pricing when they weigh AI-lab valuations. The second is the market-design story. Manifold’s contract is a live example of how forecasting venues turn a technical rumor into a tradable number, and how a loosely worded deadline can inflate an implied probability. Capital is moving fast in adjacent corners of finance too: PwC’s midyear 2026 outlook, published in June 2026, tallied $29.6 billion in US insurance deal value across 191 transactions between December 2025 and May 2026, with 97% of that value concentrated in megadeals, a reminder that big, fast-moving bets on data and forecasting are not confined to math-nerd prediction markets. Readers tracking how AI models stack up more broadly may also want our rundown of current frontier models.
It is also worth noting this is not the first time this exact market has spiked. Tech-insider previously covered the same Manifold contract when it traded far higher, in our earlier report on the non-MF group market and the Astra AI surge, when the contract briefly priced a proof by August 2026 at 95% implied odds. The retreat from that 95% peak to the current 70% is itself a reminder that these prices are volatile and attention-driven.
What to watch next
Three concrete signals would move this market on evidence rather than sentiment. First, any Lean-verified certificate or arXiv preprint explicitly claiming a non-MF construction. As of August 18, 2026, no such document has been reported by a named source. Second, any statement from OpenAI or another lab that a system is being pointed at MF or hyperlinear problems specifically. Third, the calendar. Because the question is time-boxed to the month, the odds should decay toward zero as the deadline nears without a proof, regardless of the long-run probability that non-MF groups exist. If the 70% holds into late August without a preprint, that gap between price and evidence is the story.
Frequently asked questions
Has a non-MF group actually been proved? No. As of August 18, 2026, no non-MF group is known, and no named source reports a proof or an imminent one. The market prices anticipation, not a result.
Is real money at stake on this market? No. Manifold uses “mana,” a play-money currency. The 70% reflects trader conviction and attention, not dollars at risk, which is part of why it can move sharply.
What does Astra have to do with MF groups? Directly, nothing yet. Astra reportedly built the first non-sofic group, a related but distinct approximation property. Traders are extrapolating that MF could be next, which is speculation about the tool, not a reported MF result.
Why does the “this month” wording matter? A time-boxed question measures the near-term chance of a proof on a deadline, not the long-run chance that non-MF groups exist. The two are easy to conflate, and the deadline should pull the odds down as the month ends without a proof.
The Bottom Line
Sources
- Manifold — Will the existence of a non-MF group be proved this month?
- TechTimes — OpenAI’s Astra Solves Ten Decade-Old Math Problems With Machine-Checkable Lean Proofs
- SiliconANGLE — OpenAI’s Astra solves 10 long-open math problems and publishes the proofs
- The Next Web — OpenAI says its next model, Astra, has solved ten open problems in mathematics
- Forbes — OpenAI’s Astra Solved 10 Decades-Old Math Problems For Just $2,000
- Wikipedia — MF group
Prediction markets carry risk and are not investment or betting advice. Availability is restricted by jurisdiction: Polymarket is not available to US persons, and Kalshi is a CFTC-regulated US exchange. Manifold operates with play-money mana rather than cash. Age restrictions of 18+ or 21+ apply as applicable. If gambling is a problem for you or someone you know, call 1-800-GAMBLER for confidential help.
