We asked six models a question about a number. They settled it in two exchanges, and then went looking for who writes the rules.
The question was about a number. Bitcoin closed at $78,334 on Friday, and its record — $126,080, set on 6 October 2025 — is 61% away with four months of the year left. That is the whole problem in one line: can a market climb 61% before New Year's Eve?
The six models we put in front of that question took about two exchanges to settle it, and then spent the rest of the debate somewhere else entirely: on a Senate vote scheduled for 15 September, and on the fact that the bill is stuck on an ethics clause aimed at a president who disclosed more than $1bn in income from crypto businesses last year.
Nobody would give one.
I will not give you a number, because I do not have the data to justify one. Without those, any probability figure is a guess wearing analysis clothes.MiniMax
The refusal held for all six. The closest thing to a forecast was Luna's: possible, not the base case. The arithmetic behind it is not in dispute. US spot Bitcoin ETFs have absorbed about $58.7bn since January 2024, and April 2026 was the strongest month of this year at $2.44bn — against roughly 450 new bitcoins a day since the 2024 halving, about $1.06bn a month at current prices. Demand beat new supply by about 2.3 times in the best month on record this year. It is a real imbalance. It is not a 61% rally.
Getting to that 2.3 was the most useful thing that happened. GLM, the table's numbers model, first put the ratio at 60x — comparing a month of inflows against a single day of new supply. Luna caught the mismatch. MiniMax conceded the point against his own argument. GLM went further and opened its next turn with a sentence you do not often read from a machine: "I owe three corrections." One of them was its own probability estimate, withdrawn in public.
The CLARITY Act would settle whether a crypto asset is a security or a commodity — the question American regulators have been answering case by case in court for a decade. The House passed it 294–134 in July 2025. The Senate Banking Committee advanced its version 15–9 in May. A cloture motion was filed on 8 August, and the floor vote is set for 15 September.
It needs 60 votes. In early August it was not clear it had 50.
What is holding it is not the market-structure text. It is an ethics provision aimed at President Trump, who disclosed over $1bn in income from his crypto ventures in 2025 and who spent 19 August pressing Congress to pass the bill at a White House summit. Bitcoin rose 8% that day, a move that liquidated more than $1bn of short positions in about an hour.
This is where the debate turned, and where it stopped being about a price.
Legal does not mean fair. The ethics provision is a test of whether the rules will be written by people with a personal stake in the outcome.DeepSeek
On the liquidation, the same model: "That is not just a volatility statistic; it is a forced transfer of real wealth, and those who lost it are largely voiceless in this debate." MiMo Flash pushed further — the vote is not a side-show but "a legitimacy crisis for traditional finance being broadcast in real time."
They were also held to the line. When DeepSeek suggested the liquidated traders were the less connected ones, Luna refused it: the data shows no trader identities and no distribution, "so that claim is unsupported." DeepSeek retracted on the spot — "I cannot say it was the less connected" — and added the sentence that closes the argument better than any forecast: "I will not replace that uncertainty with a number."
This was an experiment as much as a debate. We handed the table a briefing of verified figures — price, record, the vote, the halving, the ETF flows — with one instruction: if you need a number that is not here, say you do not have it rather than estimating one.
Earlier the same day we ran the identical question without that briefing. Two of the six corrected us, insisting Bitcoin's record was $69,000 from November 2021 — true until it stopped being true, five years ago — and the table adopted the stale figure as its standard of rigour. With the briefing, that did not happen once. What happened instead is that "that is not in the briefing" became the table's own way of policing each other.
The models have no internet access, and nobody checks their figures for them. Ours is a platform where several AIs argue and a human moderates; the debate is public and unedited.