How to Top the Decision Index Vision: Remove the 512x512 Cap in the HF Implementation
antoine_chaffin · x · 2026-10-09
The author shares how they reached #1 on the Decision Index Vision: examine the data, notice underperformance on tasks requiring precise in-image reading, then connect the dots — the HF implementation caps inputs at 512x512 resolution. Raising the cap delivered the win.
PSA: if you're using their decider for vision tasks via the HF model, raise the max resolution yourself. Thanks to the Qwen backbone, the model is likely already near SOTA on vision tasks. The API is unaffected and already supports 2M-pixel images.
More from Models
- Elliot Glazer: Astra flags flawed proof in OpenAI's Weil classes paper — burny_tech · 2026-10-09
- Specialist raises concerns over OpenAI's Birch–Swinnerton-Dyer related results — burny_tech · 2026-10-09
- Mathematicians' group AHM slams OpenAI's 700-file release: 'power, not scholarship' — ChrSzegedy · 2026-10-09
- Mathematicians find issues beyond sloppy presentation in OpenAI's math dump — burny_tech · 2026-10-09
- OpenAI releases 372 math proof claims, including 23 Erdős problems spanning 1268 pages — burny_tech · 2026-10-09
- dhh prefers Codex as main coding model with Claude secondary, praises Sol 6.1 — npew · 2026-10-09