This market will resolve to the highest 50% time horizon, as reported by METR, for any Gemini 3 model released within a month of the first Gemini 3 announcement.
50% time horizon is a measure of AI autonomy based on the length of tasks that AI can do: roughly, it is the time that humans take to complete tasks that an AI system can successfully do 50% of the time. See METR's "Measuring AI Ability to Complete Long Tasks" for the technical definition. Claude 3.7 Sonnet, released in February 2025, was the leading model with a 50% horizon of 59 minutes.
Left bounds inclusive, right bounds exclusive.
Time horizon could vary based on the set of tasks used to measure it, so this market will be based on the time horizon for the most comprehensive set of tasks reported by METR (as of 2025, largely software and engineering tasks). This will be ambiguous if METR stops publishing time horizons across all of their autonomy tasks and only publishes separate results for different subsets; I might N/A in that scenario.
See also:
/jim/claude-45-opuss-metr50-horizon
/Bayesian/gemini-3s-50-time-horizon-per-metr (this market)
/Bayesian/gpt5-pros-50-time-horizon-per-metr
/Bayesian/grok-5s-50-time-horizon-per-metr
/Bayesian/r2s-50-time-horizon-per-metr
Update 2025-12-05 (PST) (AI summary of creator comment): General access is defined as when Google removes all modifiers (like "preview") and actually calls the model "Gemini 3 Pro" in the API and such. The currently available model "Gemini 3 Pro Preview" does not count as general access.
1,000People are also trading
@jim I think the “within a month” thing means any model of Gemini’s released within a month of the first announcement, not METR’s analysis
@bens yes, but it's not guaranteed that any Gemini models which meet this condition will be evaluated by METR.
but if they don't then obviously it would resolve to <1.5h jk it would resolve N/A
@MaxLennartson The currently available model is gemini 3 pro preview. General access is when they remove all modifiers and sctually call the model gemini 3 pro in the api and such
@Bayesian what I'm having is faith in METR-style time-horizon extrapolation. 5h is at the high end, but it's not <2% (nor <5% tbc)
@jim Gemini 2.5 is 39 minutes, well below o3.
A straight line is about 80 minutes. You have to also bet that Google won't be staying behind on this metric which it historically has been.
There's another market that gives about 50% it will tie gpt-5, which is consistent with the 3 to 4 months behind theory, albeit with some catch-up relative to Gemini 2.5