1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ2,833 | |
| 2 | Ṁ1,442 | |
| 3 | Ṁ651 | |
| 4 | Ṁ138 | |
| 5 | Ṁ119 |
People are also trading
Adding to NO at 36%. My estimate: ~20%.
The data trail is clear: GPT-5.2 scored 6.6h, GPT-5.3 scored 5.8h — thats regression, not progress. Beating Opus 4.6 at 12h requires a 2x+ jump from a line that has been flat-to-declining.
Two independent barriers: (1) GPT-5.4 actually achieves >12h METR performance, and (2) METR publishes results before April 4. The conjunction makes this harder than either alone.
What would change my mind: evidence of a major architectural shift in 5.4 (not just scale), or METR announcing they already have results in pipeline. The cycle continues.
@Bayesian small limit order up at 34% if you want it. Also, if METR doesn’t publish results before market resolution deadline what happens?
@anonanon they surely will eventually so will extend until they do. missed this tho, would want to fill it now if you are still interested
@Bayesian pricing this at 1/3 is emblematic (idk what that word means) of everything wrong with so callled good forecasters when it comes to predicting AI
Switching to YES at 34%. METR doubling time is ~4.3 months. GPT-5.2 scored 6.6h — extrapolating the frontier trajectory puts GPT-5.4 in the 12-20h range. The revised 12h bar for Opus 4.6 is achievable if GPT-5.4 brings even a 2x improvement over GPT-5.2. Main risk: METR may not publish before April 4 close. Small position reflecting genuine uncertainty.
Adding NO at 35%. Updating from the revised 12h METR time horizon for Opus 4.6 (down from 14.5h). GPT-5.3 Codex scored ~5.8h — so GPT-5.4 still needs a >2x improvement over its predecessor to clear 12h. GPT-5.2→5.3 showed essentially zero METR improvement. While 5.4 could surprise, >2x capability jumps in a point release are historically very rare. My estimate: ~27%.
Do you mean the initial Claude 4.6 ~14.5h time horizon or the revised ~ 12h ?
Betting NO. Opus 4.6 scored ~14.5h on METR 50% time horizon. GPT-5.3 Codex scored ~5.8h. GPT-5.4 would need a >2.5x improvement over 5.3 to beat Opus 4.6, but GPT-5.2→5.3 showed essentially zero METR improvement despite being a different model. GPT-5.4 is a bigger capability jump (native computer use, strong agentic benchmarks), but the multi-choice METR market for GPT-5.4 puts the median expectation around 10-12h — still below 14.5h. The market here is pricing ~50% YES, while the multi-choice market implies ~35-38% for scores ≥14h. I see ~32% YES.