Image & vision

SixthView

by Pooya Khosravi, Jen-Feng Chang and Fnu Pramono (repository under facebookresearch)

What is SixthView? SixthView is an image & vision eval made by Pooya Khosravi, Jen-Feng Chang and Fnu Pramono (repository under facebookresearch). A spatial-reasoning benchmark. The model gets five orthographic views of a block built from interlocking cube pieces, plus information about the pieces, and must predict the sixth, hidden view as a grid of part labels, scored both on the exact grid and cell by cell. The paper (arXiv:2611.23891) describes 602 synthetic assemblies of 2 to 10 parts and checks each item so that only one hidden view fits the inputs; the repository's data files list 604 rows. Inputs come as text only, as images, or both. The authors report that three vision-language models reached 33.2% to 85.4% exact-grid accuracy with images plus exact text, and that adding exact text helped far more than adding images.

Page last updated 2026-10-18 · Evals Review

Editor note

A spatial-reasoning benchmark. The model gets five orthographic views of a block built from interlocking cube pieces, plus information about the pieces, and must predict the sixth, hidden view as a grid of part labels, scored both on the exact grid and cell by cell. The paper (arXiv:2611.23891) describes 602 synthetic assemblies of 2 to 10 parts and checks each item so that only one hidden view fits the inputs; the repository's data files list 604 rows. Inputs come as text only, as images, or both. The authors report that three vision-language models reached 33.2% to 85.4% exact-grid accuracy with images plus exact text, and that adding exact text helped far more than adding images.

What it tests
3D assembly and occlusion reasoning from multiple views, with and without exact symbolic geometry.
Original source
github.com/facebookresearch/sixthview_benchmark
Last updated
arXiv:2611.23891 (18 Nov 2026); repo last push 2026-10-29 (GitHub API, read 19 Nov 2026)
License
CC BY-NC 4.0, non-commercial (repo LICENSE file, read 19 Nov 2026)

Facts come only from the linked source (checked 15–16 Oct 2026). Credit belongs to the creators. We don't score or rank evals: how we review.

Reviews

No reviews yet.

Reviews are written by people who sign in with GitHub, one per account per eval. We can't confirm that each reviewer used the eval; we remove reviews that break our rules, never just for being negative, and we never add, buy or edit reviews. Averages, when shown, come only from these reviews.

Sign in with GitHub to review

Related image & vision evals

  • MMMU — Expert-level multimodal understanding and reasoning.
  • MathVista — Mathematical reasoning over figures, plots and images.
  • MMBench — Broad perception and reasoning skills of vision-language models.
  • ChartQA — Question answering over charts.

All Image & vision evals · Comparison table

Discussion

No one has discussed SixthView yet. Have you run it? Share what it caught, what it missed or how you set it up.

Sign in with GitHub to comment

Be specific and kind. No promotion or affiliate links. Community rules · All discussions