Axis Spatial

Search modelsMenu

Updates

New models and results, releases and corrections, newest first. Follow them with the Atom feed or by email.

  1. · Release

    Release 1.3

    • A page for each lab: every model we benchmark from it, on every evaluation.
    • The home page is now an overview, and the Geospatial Intelligence Index has its own page.
    • Comparison pages open with the answer, and every page has a model search.
    • A simpler phone header with a grouped menu.

  2. · Correction

    Nemotron 3.5 Super VL 120B: Geospatial knowledge score corrected

    • Its Geospatial knowledge score is 92, not 89. The provider returned empty replies to some questions, which had been graded as wrong; they were asked again.

  3. · Release

    Release 1.2

    • A small chart for each evaluation on every model page.
    • Comparisons open on Anthropic against OpenAI, tier for tier, and follow each family's newest version.
    • Safety and Geospatial knowledge results are given as rates.

  4. · Release

    Release 1.1

    • A Tasks page with a world map of where the tasks are set.
    • A Safety page: how often each entry tried to game the checks.
    • Pages for six more evaluations, among them OGD4All, GeoChrono and BigEarthNet.txt.
    • Every developer and every model family on every chart.

  5. · New results

    First results for Claude Opus 5.5 (low), Claude Sonnet 5.5 (low) and Claude Haiku 5.5 (low)

  6. · New results

    First results for MiniMax M3.1 Flash (max), Nemotron 3 Nano Omni 30B (Reasoning) and 3 more

  7. · New results

    First results for Nemotron 3 Ultra (high), DeepSeek V4.1 Flash (max) and Qwen 3.8 27B (xhigh)

  8. · Methodology

    Version 1.0 of the methodology

    • First public release.
    • The method is the same as in Pre-release 0.34: no task, weight or scoring rule changed.
    • Version history