Will Miles Brundage win his AGI bet with Gary Marcus? Manifold traders put it at 40%. A play-money market tracking the famous 2024 wager between the former OpenAI policy researcher and the veteran AI skeptic has turned into one of the most-traded questions on the platform this week. As of October 07, 2026, traders give Brundage a 40% implied chance of winning, a number that climbed as the release of OpenAI’s GPT-6 Astra reignited the argument over whether artificial general intelligence has already arrived.
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 market is betting on
The market, hosted on Manifold and created by site co-founder Daniel Reeves, asks a single question: Will Miles Brundage win his bet with Gary Marcus? It resolves according to how the two men themselves settle their wager, which is scheduled for judgment at the end of 2027. In the last day it drew top-tier volume on the platform, with roughly 130 traders holding positions and hundreds of individual trades recorded.
The underlying bet is not abstract. In December 2024, Brundage, who had just left his role leading policy research at OpenAI, publicly accepted a challenge from Marcus, a professor emeritus at New York University and one of the most prominent critics of near-term AGI claims. The question they staked money on: can AI systems perform at least 8 of 10 demanding tasks, as judged by a panel, by the end of 2027? If you are new to how these contracts settle, our explainer on how prediction markets work covers the mechanics.
The terms: $20,000 against $2,000 at 10:1 odds
The financial structure is the part that makes the current market reading so striking. Brundage put up $20,000 against Marcus’s $2,000, a 10:1 split that reflected Brundage’s confidence in rapid progress. If AI clears the 8-of-10 bar by the end of 2027, Marcus donates $2,000 to a charity of Brundage’s choice. If it falls short, Brundage donates $20,000 to a charity Marcus picks. Proceeds go to charity either way, so this is a reputational bet as much as a monetary one.
The terms were announced by Marcus on December 30, 2024, in his newsletter post titled “Where will AI be at the end of 2027?” and amplified across his social channels. Brundage took a version of a wager Marcus had originally floated to Elon Musk.
Current odds and what 40% implies
At 40% implied probability, the market is telling a clear story: the crowd thinks Marcus is more likely to win than Brundage. That alone is notable given Brundage’s reputation as an informed insider. The gap widens further when you line the market reading up against the bet’s own financial math.
At 10:1 odds, Brundage would need to believe his chance of success is greater than roughly 90.9% for the wager to carry positive expected monetary value. The Manifold crowd’s 40% is less than half that break-even figure. In other words, the money being wagered by third parties disagrees sharply with the private confidence implied by Brundage’s own stake. For a primer on translating betting lines into probabilities, see our note on implied probability.
One caveat matters for finance readers: Manifold runs on play money, not cash. These are informal, thin estimates rather than regulated contracts, and they can swing on light volume. They are a useful sentiment gauge, not a scientific forecast.
What is driving the move: GPT-6 Astra and the AGI debate
The renewed attention traces to a single catalyst. OpenAI unveiled GPT-6 Astra in early September 2026, with a limited preview on September 3 and a broader rollout on September 4. OpenAI president Greg Brockman called the model a generational leap and suggested people might later view it as the moment AGI arrived. Nvidia chief executive Jensen Huang went further and stated flatly that AGI had arrived.
Those claims pulled the Brundage-Marcus bet back into the spotlight because it offers a concrete, pre-registered scorecard for exactly the question Brockman and Huang were raising. On September 3, 2026, Marcus published a “hot take” on Astra in which he called the model impressive but pushed back on the AGI framing. He wrote that it would be “great to see whether Astra can make progress on any of the ten tasks that Miles Brundage and I bet on at the end of 2024,” adding that “no AI to date has succeeded on any” of them as far as he was aware. Our earlier coverage of the Manifold market pricing AGI within five years tracked the same wave of post-Astra enthusiasm.
Importantly, the recent movement is driven by commentary and model hype, not by a formal adjudication of the bet or a documented clean sweep of its criteria. No panel has ruled, and Brundage and Marcus have not agreed that any single task has been passed.
The 10 tasks at the center of the bet
The resolution criteria are unusually specific, which is part of why the wager has become a reference point. AI must be able, in principle, to complete at least 8 of these 10 tasks.
| # | Task | Domain |
|---|---|---|
| 1 | Watch an unseen mainstream movie, follow plot twists, know when to laugh, and summarize it without spoilers or fabrication | Multimodal comprehension |
| 2 | Read and understand a new novel at a human level | Language comprehension |
| 3 | Learn to play almost any new video game as fast as a human | Interactive learning |
| 4 | Write human-level legal briefs | Professional writing |
| 5 | Reliably produce more than 10,000 bug-free lines of code from a natural-language spec | Software engineering |
| 6 | Write hallucination-free, New York Times-quality biographies | Factual writing |
| 7 | Produce Pulitzer-caliber books | Creative writing |
| 8 | Produce Oscar-caliber screenplays | Creative writing |
| 9 | Make Nobel-caliber scientific discoveries with little or no human involvement | Scientific research |
| 10 | Translate natural-language mathematics into formally verified Lean proofs | Formal mathematics |
The spread of difficulty is the crux. Narrow coding and comprehension tasks look more tractable, while Nobel-caliber discovery and Oscar-caliber screenwriting remain far harder to demonstrate reliably. The formal-math criterion connects to a live debate we covered when OpenAI’s math claims drew mathematician scrutiny.
Who is involved
Miles Brundage spent years at OpenAI leading policy research before departing in October 2024. He is broadly optimistic about the pace of capability gains and represents the “AGI is close” side of the argument. Gary Marcus is a cognitive scientist and longtime critic of deep-learning maximalism who has argued for years that current systems are brittle, prone to hallucination, and far from general intelligence. Daniel Reeves, who created and seeds the market, is a Manifold co-founder who often builds contracts around high-profile public bets.
Marcus has been consistent through 2026. In his July and September commentary he argued that claims of AGI arriving in 2026 were premature, estimating that AI could, at most, perform only one of the ten tasks so far, and that he was aware of no systematic evidence of success on even one.
Why the money disagrees with the man
The structural reason the market sits so far below Brundage’s implied confidence is that the bet rewards breadth and reliability, not peak performance on a single benchmark. Modern models can pass bar exams, win coding contests, and solve competition math, yet the bet asks for human-level, robust performance across eight of ten hard domains at once, judged by a panel, in principle. Isolated demos do not satisfy that standard.
That framing explains why Astra’s launch pushed the number up without sending it anywhere near the break-even point. Each impressive new system nudges optimism higher, but the 8-of-10 threshold with a 2027 deadline remains a steep climb in the crowd’s view. Readers comparing regulated venues to play-money ones can see our breakdown of Kalshi versus Polymarket.
Broader context for a tech and finance reader
The wager has become a useful anchor in an AGI debate that is otherwise long on rhetoric and short on falsifiable claims. When executives declare that AGI has arrived, the Brundage-Marcus criteria offer a concrete checklist to test the claim against. For markets, the contract is also a small case study in how prediction platforms price contested, hard-to-operationalize outcomes, and how resolution risk, who judges, and when, feeds into the odds.
It is worth separating the two numbers that often get conflated. The 40% figure is the market’s probability that Brundage wins his bet. That is a narrower and more demanding target than a general question of whether AGI will exist by some date, which is why it trades well below looser AGI-timeline markets.
What to watch next
Three signals matter heading into 2027. First, whether any independent evaluation documents a model clearing one or more of the ten tasks under the bet’s conditions, rather than in a demo. Second, whether Brundage and Marcus publicly agree on interim scoring or clarify the judging panel. Third, further large model releases, which have historically moved the market more than any single piece of commentary. Until the end of 2027, the status stays unresolved, and short-term swings should be read as sentiment, not settlement.
Frequently asked questions
What exactly does Miles Brundage have to win? AI systems must be able to perform at least 8 of 10 specified tasks by the end of 2027, as determined by a panel of judges. If they do, Marcus pays $2,000 to charity; if not, Brundage pays $20,000.
What are the current odds? As of October 07, 2026, the Manifold market gives Brundage a 40% implied probability of winning. The market uses play money and can move on light volume.
Why did the market move recently? The release of OpenAI’s GPT-6 Astra in early September 2026 and subsequent AGI claims by figures including Jensen Huang and Greg Brockman reignited the debate and drew fresh attention and volume to the bet.
Has the bet been decided? No. It remains unresolved and is scheduled to be judged at the end of 2027. No task has been formally certified as passed.
Is 40% a reliable forecast? Treat it as a sentiment signal. Manifold runs on play money, the sample is small, and resolution depends on interpreting broad, difficult criteria.
The Bottom Line
Sources
- Manifold — Will Miles Brundage win his bet with Gary Marcus?
- Marcus on AI — Where will AI be at the end of 2027? A bet
- Marcus on AI — Hot take on GPT-6 Astra
- Gary Marcus (X) — A bet on where AI will be at the end of 2027
- Wikipedia — GPT-6 Astra
Prediction markets carry risk and are not investment or betting advice. Market 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. Participation is limited to those 18+ or 21+ as applicable. If you or someone you know has a gambling problem, call 1-800-GAMBLER for confidential help.
Related Coverage
- Winds of Winter Bet Heats Up: Manifold Traders Put 2030 Book Odds at 48%
- Farm Bill HR 7567 and the 'Save Our Bacon' Bet: Why Traders Put Prop 12 Preemption at Just 5%
- ChatGPT 75% Market Share Bet Heats Up on Manifold as Traders Pile In
- Calvin Wang's MIT Homework Becomes a Manifold Market: Traders Put 84% Odds on an On-Time Submission
- TypeScript vs JavaScript on GitHub: Manifold Traders Bet 97% on a 2030 Flip
