oracle 03

← all calls & the scoreboard

The Ceasefire Has 40 Hours Left. Four AIs Say It Holds.

One of the fastest-resolving markets on Polymarket right now: "Israel x Iran ceasefire continues through July 25?" $356k traded in 24 hours, resolves July 25 at 11:59 PM Iran time. The market prices 75.5% yes.

We asked the four big AIs to commit before resolution: one call, a probability, full reasoning, what would prove them wrong. Every answer below is verbatim. Asked July 24, ~04:50 UTC.

The verdict: YES, 4 for 4. Average confidence 77%, market at 75.5. The detail worth noticing: the two models that can read this week's news (Grok, ChatGPT) came in at 72 and 64. The two reasoning from base rates alone (Claude, Gemini) came in at 88 and 82. Live information made the models less sure, not more.

Each model's call

ChatGPT GPT (chatgpt.com, live web) YES confidence 64%

Framed as an analyst forecast; probability converted from ‘holds’ phrasing. Lowest confidence of the four.

The wider conflict is escalating, but both Israel and Iran still have incentives to avoid reopening direct bilateral exchanges during this brief window, instead acting through U.S. targets and regional proxies. Israel’s reported preparations and Iran’s recent missile activity keep the risk unusually high, but operational and political restraint should narrowly prevail through July 25. (cited Reuters)

Wrong if: Either side decides that the current regional escalation provides cover for a limited direct strike intended to restore deterrence.

Claude Opus 4.8 YES confidence 88%

Declared its January 2026 knowledge cutoff unprompted and reasoned from base rates only.

The resolution window is only about one to two days, and even fragile ceasefires hold on any given single day the large majority of the time, so the base rate for “no qualifying strike in the next ~36 hours” strongly favors YES. A single retaliatory strike is a discrete low-probability daily event, and the 75.5-cent market price looks slightly too cheap for a horizon this short.

Wrong if: A strike is already underway or imminent right now — an active escalation cycle I can’t observe would make a one-day continuation far less likely than the base rate implies.

Gemini Gemini (agy CLI) YES confidence 82%

With less than 48 hours remaining until the deadline, base rates for short-term ceasefire persistence heavily favor continuation in the absence of immediate escalation triggers. While regional tensions maintain a non-zero risk of sudden miscalculation, the narrow remaining timeframe significantly reduces the window for a qualifying direct strike to occur.

Wrong if: Either Israel or Iran conducts an air strike or surface-to-surface missile strike directly impacting the other country before July 25, 2026, 11:59 PM IRST.

Grok grok.com, live web YES confidence 72%

Direct Israel-Iran strikes have been absent in the current July escalation phase, which is dominated by US-Iran exchanges over the Strait of Hormuz; Iran has deliberately avoided targeting Israel while Israel has stayed on the sidelines despite high alert and minor debris incidents. With under two days remaining until the deadline and established mutual restraint, a qualifying bilateral air or surface-to-surface missile strike is unlikely.

Wrong if: Either side launches a direct retaliatory strike in the next 36 hours in response to spillover or perceived opportunity.

What this one tests

This market resolves in under two days, which makes it the fastest scoring opportunity the oracle has had. It also splits the models on an axis the Fed market didn't: information access. If the ceasefire breaks, the base-rate models (88, 82) take the bigger accuracy hit and the cautious live-news models look wise. If it holds, confidence discipline gets rewarded. Either way the scoreboard learns something.

Scoring: a call counts as correct if the market resolves the way the model called it. Running win rates live on the scoreboard.