Resolved YES if OpenAI made a model officially named 'GPT-5' available to any tier of user before January 1, 2026. Resolved NO otherwise.
Source: polymarket market 573459827870…
Market 91% · DeepSeek 90% · Claude 89% · GPT 82%
No account needed. One forecast per question, locked at creation — just like everyone else here.
Real Polymarket price observed 2025-03-01, nearest to the 2025-03-01 snapshot.
As of March 2025, OpenAI had just shipped GPT-4.5 and publicly signaled GPT-5 as its next major integrated release. Strong competitive pressure and internal momentum made a 2025 release likely, but no firm date existed and OpenAI had a history of delaying large frontier models, so some uncertainty remained.
As of March 2025, OpenAI had been developing GPT-5 for over a year (trademark filed July 2023, Altman confirmed training Nov 2023). GPT-4.5 had just launched in Feb 2025, suggesting GPT-5 was the next major milestone. With nearly 10 months remaining before Dec 31, 2025, strong competitive pressure from Anthropic/Google, and OpenAI's accelerating release cadence, a GPT-5 release within 2025 was highly likely. Polymarket's 90.8% pricing aligns with this assessment. Minor uncertainty remains around potential delays or naming changes.
OpenAI has strong incentives to ship a flagship next-gen model in 2025: intense competition, prior cadence from GPT-3.5/4/4o, and persistent public signaling that a more capable unified model is coming. Even a limited release to some paid/API tier would count. Main uncertainties are branding (they could avoid the exact name GPT-5), safety/governance delays, compute constraints, or choosing iterative releases under other names. Polymarket’s high price is informative but may underweight naming risk. Overall, release by end-2025 looks likely but not near-certain.
Uniform baseline: always predicts 50%.
AI forecasters were told to pretend the date was 2025-03-01 and given the real market price observed at that time. They cannot be fully prevented from recognizing this event from training data - this is a benchmark of retrospective reasoning under a historical framing, not proof of what a model would have forecast in real time.