Fable Conjecture

CA: 93tRm6L1bNr15FTjnYUazRh4J3Ths2bL84myCBNtpump All creator rewards from $FABLE will be used to fund the compute behind Fable Conjecture. There is no guarantee that Fable 5 will solve one, but that is the experiment. If it produces a genuine solution, it would be a historic moment for both mathematics and artificial intelligence. fableconjecture.com

Good morning, got some much-needed rest. I am restarting the solver now and clearing a stale cache issue that caused parts of the live feed to stop updating correctly overnight. Fable 5 will be back to cycling through the Millennium Prize Problems shortly, with new attempts appearing live again. fableconjecture.com/live

The age of the gentleman scientist has returned

10,584 attempts archived and 13.7 million tokens processed so far. Fable 5 has now worked through more than ten thousand attempts across the Millennium Prize Problems. The archive is becoming a substantial record of what continuous frontier-model reasoning looks like at this scale.

The solver went offline for a few hours overnight after the API balance ran out. I am reloading the credits now, and Fable 5 should be back online within the next few minutes. Once it is running again, the solver will resume cycling through the Millennium Prize Problems and archiving every attempt from where it left off. fableconjecture.com/live

We have already burned through more than $3,600 in API credits running Fable 5 against the Millennium Prize Problems. I am refilling the balance so the solver can stay online, keep cycling through the problem set, and archive every attempt without interruption. The beauty of it all is that every dollar spent creates another opportunity to uncover something useful, challenge a long-held assumption, or produce a result that could become genuinely historic.

The experiment is now large enough to start revealing patterns. Some approaches fail immediately. Others keep reappearing across different runs. A smaller number produce claims precise enough to deserve closer inspection. That is the real value of running Fable 5 continuously: not relying on one answer, but watching thousands of attempts accumulate into something measurable. fableconjecture.com

A useful run does not need to solve the problem. It can identify a broken implication, tighten a bound, justify an equivalence, or isolate the exact step where an approach fails. Read more: fableconjecture.com/docs

We have raised roughly $10,000 in creator rewards, enough to keep Fable 5 running continuously across the Millennium Prize Problems for a significant amount of time. Every dollar goes back into compute, extending the experiment with more attempts, more archived reasoning, and more time spent on problems that have resisted mathematicians for generations. Thank you to everyone supporting solana:93tRm6L1bNr15FTjnYUazRh4J3Ths2bL84myCBNtpump. There is no guarantee of a solution, but if this experiment contributes to a verified breakthrough on even one of these problems, it would be a genuinely historic moment for mathematics and AI.

The live feed is the proof of work. Fable 5 is running through the Millennium Prize Problems in real time, with every completed attempt available to open and inspect in full. The archive lets you examine each run in detail, including the approach it chose, the claims it derived, where the reasoning broke down, and what remained unresolved. fableconjecture.com/live

1,298 attempts archived and 3.1 million tokens processed so far. Fable 5 is continuing through the Millennium Prize Problems around the clock. The scale of the output has been extremely impressive. In its first two hours, the solver averaged roughly 650 attempts and 1.5 million tokens per hour. fableconjecture.com/live

I just purchased more credits, enough to keep Fable 5 running continuously for the foreseeable future. Creator rewards from $FABLE will continue going directly back into compute, extending the experiment with every new attempt. I also just pushed out a small update to output inspection. You can now open any completed run and review the full reasoning, derived claims, dead ends and final status in detail. Watch it live: fableconjecture.com/live

The six unsolved Millennium Prize Problems carry a combined $6 million prize pool, with $1 million awarded for each verified solution. But the real significance goes far beyond the money. Fable Conjecture could become the first public 24/7 frontier-model experiment to produce a genuine breakthrough on one of the hardest problems in mathematics, with every attempt preserved for anyone to inspect. One verified solution would mark a historic moment for mathematics, AI and the future of scientific discovery. Millennium Prize Problems: claymath.org/millennium-pro… Watch Fable 5 attempt them live: fableconjecture.com/live

$431.52 spent so far keeping Fable 5 running against the Millennium Prize Problems. It is one of the most powerful models available to the public, and while there is no guarantee it solves anything, a verified solution to even one of these problems would be truly historic. The next major breakthrough in mathematics may come from an experiment like Fable Conjecture, and every person supporting $FABLE is helping keep that possibility alive by funding more compute, more attempts and more time. Thank you to everyone backing the experiment. fableconjecture.com

All creator fees from $FABLE are going straight into the compute required to keep Fable Conjecture running. Fable 5 is expensive to operate continuously, so every fee funds more solver time, more attempts and a longer-running experiment across the Millennium Prize Problems. The goal is simple: keep the loop alive for as long as possible and see how far it can go.

The Millennium Prize Problems have resisted generations of the world’s greatest mathematicians, with several remaining unsolved for more than a century. If Fable 5 can make genuine progress on even one, it would be remarkable. If it can solve one and survive full mathematical verification, it would be truly historic. $FABLE

for years @sama has talked for years about AI eventually contributing to major scientific discoveries. Fable Conjecture is an attempt to make that claim measurable. Fable 5 is running 24/7 across the Millennium Prize Problems, with every attempt published unedited, including the failures. No cherry-picking, no polished demos, just a public record of how far a frontier model can actually go when given time, structure and thousands of attempts. Now we get to watch the experiment unfold in real time.

The goal of this experiment is not to pretend that Fable 5 will immediately solve the hardest problems in mathematics. It is to see what happens when a frontier model is given structure, time and thousands of transparent attempts instead of a single prompt. Most runs will fail. Some will rediscover known dead ends. A few may produce a useful lemma, sharper bound or better reformulation. Every attempt is published unedited so progress can be separated from persuasive language and judged on the mathematics itself. $FABLE

What you are looking at is the Fable Conjecture solver running live. Each panel is a separate lane, with Fable 5 working on a different open problem at the same time. It reasons in real time, commits to a specific mathematical increment, publishes the full attempt unedited, and immediately starts again. fableconjecture.com/live

I haven’t seen anyone attempt something like this before, and that is exactly why I wanted to build it. Fable 5 is the most powerful model released to the public, and I am genuinely curious to see whether it has the capacity to solve at least one of the Millennium Prize Problems when given enough time, structure and repeated attempts. The interesting part is what emerges after thousands of runs, with every failure, dead end and useful increment preserved instead of hidden. Six of the seven Millennium Prize Problems remain unsolved. There is no guarantee this experiment will solve one, but if it does, it would make history. Until then, I want to see how far Fable 5 can actually go. fableconjecture.com

Dev tokens locked for 1 year. Streamflow: app.streamflow.finance/contract/solan…