This market will resolve to the highest 50% time horizon, as reported by METR, for the first GPT-5.4 model to appear on METR's graph. only GPT 5.4 counts, otherwise N/A.
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.
See also:
/Bayesian/gpt-52-pro-metr-time-horizon
/Bayesian/gemini-3s-50-time-horizon-per-metr
/Bayesian/gemini-3-pro-metr-50-time-horizon
/Bayesian/claude-sonnet-46s-metr-50-time-hori
/Bayesian/claude-sonnet-5-metr-50-time-horizo (this market)
/Bayesian/claude-opus-5-metr-50-time-horizon
/Bayesian/grok-420s-metr-50-time-horizon
/Bayesian/grok-5s-50-time-horizon-per-metr
/Bayesian/r2s-50-time-horizon-per-metr
/Bayesian/kimi-k3-thinkings-metr-50-time-hori
Update 2026-04-06 (PST) (AI summary of creator comment): If METR does not report a time horizon for GPT-5.4 and seems intent on not reporting it in the foreseeable future, this market resolves N/A.
1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ1,080 | |
| 2 | Ṁ205 | |
| 3 | Ṁ172 | |
| 4 | Ṁ147 | |
| 5 | Ṁ115 |
People are also trading
Buying YES on 12-14h. The ECI-based estimates from the EA Forum (Charles Dillon) predicted Opus 4.6 at ~9-10h, but actual METR result came in at 14.5h — the model systematically underestimates by ~50%. GPT-5.3 Codex was estimated at ~8.5h by the same model, suggesting actual performance around 12-13h. GPT-5.4 should beat GPT-5.3 but I think Bayesian is right that it does worse than Opus. The 32% market mass on <10h seems too high given how much ECI predictions undershot Opus.
@jim it’s barely been a month since 5.3-codex, which was only ~6 hrs. Even if that’s a modest underestimate, 5.4 could easily be only say a 33% improvement from 7.5 - - 10 hrs. That would still be faster growth than the recent trend depending on how you look at it
@DavidHiggs that result was sussy, codex models are a different breed, have never performed well on metr. Also it wasn't tested through the API.
Also it scored like worse than GPT-5.2 which was a December model so