Benchmark Almanac

KilometerVision

CloudAI-evolve 90.5% tops its leaderboard
Not in model cards

Models watch walking-tour videos geolocated over up to 1 km and answer questions on landmarks, routes and map layout, such as distances and loop closures.

A new track of the fourth Perception Test Challenge at ECCV 2026, which opened on June 20th, 2026. The paper followed on September 30th, 2026. A sibling track, KilometerAudio, uses hour-long walking tours with audio.

Released
June 2026
Kind
Spatial reasoning and counting
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
Public, plus a private test set

In the authors’ words

We push the frontier of large-scale spatial intelligence in Vision-Language Models (VLMs) and introduce the first benchmark that probes geographical layout understanding from real-world videos, spanning up to 1km distances. Inspired by the cognitive science literature, we evaluate models against the hierarchical stages of human spatial awareness: anchoring via landmarks, connecting them through routes, and integrating these into global mental maps. Extensive experiments reveal a fundamental divergence in how current AI models process spatial information. Instead of utilising true path integration or forming geometric survey knowledge, we find that VLMs rely almost entirely on 2D visual recognition and text-matching to bypass complex spatial reasoning. The benchmark is publicly available at https://perception-test-challenge.github.io/kilometervision.html.

Abstract of KilometerVision: A New Frontier for Large-Scale Spatial Intelligence in VLMs (arXiv:2609.39588, CC0)
26
Entries on its own leaderboard, 0 models we track

CloudAI-evolve tops the leaderboard with 90.5%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

26 entries. Scroll sideways for the rest.

  1. 90.5 CloudAI-evolve
  2. 75.3 ohmyyuan
  3. 72.3 Huawei_SpaceMind
  4. 68.7 SNU CVLAB
  5. 67.2 OpenDL / entry 5
  6. 62.2 percept / adv
  7. 61.8 Banana Bats
  8. 61.6 amanupg / run-v4
  9. 60.5 Wind_Rain_Tower
  10. 58.4 amanupg2
  11. 57.7 StarAtNyte
  12. 57.6 Orpheon
  13. 56.6 AGI / AGI-KV-8
  14. 48.8 Lostagain
  15. 45.4 BoundedAccord / bounded-geometry-qwen-zero-shot
  16. 45.1 qianlan
  17. 42.7 TamakiLab / KilometerVision Specialist Routing Best
  18. 40.9 Openthu / Qwen3-VL-32B parent plus strict temporal graph specialists
  19. 40.9 ZeroOne / Qwen3-VL-32B parent plus Qwen8 distance CLIP and strict temporal graph
  20. 40.8 Math / Qwen3-VL-32B parent with strict temporal graph specialists
  21. 38.8 LiouMIT / Qwen3-VL-32B three-expert consensus
  22. 33.1 redde / redde-v7
  23. 32.6 cbb
  24. 27.7 nexus_xxy / qwen
  25. 21.5 FZJ_IAS8
  26. 19.6 Host_37595_Team / zero-shot random baseline
  1. 1 CloudAI-evolve zero-shot; submission 579751 90.5%
  2. 2 ohmyyuan zero-shot; submission 579939 75.3%
  3. 3 Huawei_SpaceMind zero-shot; submission 580913 72.3%
  4. 4 SNU CVLAB zero-shot; submission 581278 68.7%
  5. 5 OpenDL / entry 5 zero-shot; submission 580633 67.2%
  6. 6 percept / adv fine-tuned; submission 581291 62.2%
  7. 7 Banana Bats zero-shot; submission 581269 61.8%
  8. 8 amanupg / run-v4 submission 581125 61.6%
  9. 9 Wind_Rain_Tower zero-shot; submission 580990 60.5%
  10. 10 amanupg2 zero-shot; submission 581251 58.4%
  11. 11 StarAtNyte submission 580781 57.7%
  12. 12 Orpheon zero-shot; submission 580466 57.6%
  13. 13 AGI / AGI-KV-8 zero-shot; submission 580927 56.6%
  14. 14 Lostagain zero-shot; submission 581249 48.8%
  15. 15 BoundedAccord / bounded-geometry-qwen-zero-shot zero-shot; submission 579341 45.4%
  16. 16 qianlan zero-shot; submission 580716 45.1%
  17. 17 TamakiLab / KilometerVision Specialist Routing Best zero-shot; submission 581246 42.7%
  18. 18 Openthu / Qwen3-VL-32B parent plus strict temporal graph specialists 0-shot; submission 579930 40.9%
  19. 19 ZeroOne / Qwen3-VL-32B parent plus Qwen8 distance CLIP and strict temporal graph 0-shot; submission 579937 40.9%
  20. 20 Math / Qwen3-VL-32B parent with strict temporal graph specialists 0-shot; submission 579938 40.8%
  21. 21 LiouMIT / Qwen3-VL-32B three-expert consensus 0-shot; submission 579904 38.8%
  22. 22 redde / redde-v7 zero-shot; submission 580670 33.1%
  23. 23 cbb zero-shot; submission 581032 32.6%
  24. 24 nexus_xxy / qwen zero-shot; submission 581018 27.7%
  25. 25 FZJ_IAS8 fine-tuned; submission 579743 21.5%
  26. 26 Host_37595_Team / zero-shot random baseline zero-shot; submission 578763 19.6%

Source: Official KilometerVision leaderboard

#EntryScoreSettingDateSource
1 CloudAI-evolve 90.5% zero-shot; submission 579751, CloudAI-evolve Aug 2026 Official leaderboard
2 ohmyyuan 75.3% zero-shot; submission 579939, ohmyyuan Aug 2026 Official leaderboard
3 Huawei_SpaceMind 72.3% zero-shot; submission 580913, Huawei_SpaceMind Aug 2026 Official leaderboard
4 SNU CVLAB 68.7% zero-shot; submission 581278, SNU CVLAB Aug 2026 Official leaderboard
5 OpenDL / entry 5 67.2% zero-shot; submission 580633, OpenDL Aug 2026 Official leaderboard
6 percept / adv 62.2% fine-tuned; submission 581291, percept Aug 2026 Official leaderboard
7 Banana Bats 61.8% zero-shot; submission 581269, Banana Bats Aug 2026 Official leaderboard
8 amanupg / run-v4 61.6% submission 581125, amanupg Aug 2026 Official leaderboard
9 Wind_Rain_Tower 60.5% zero-shot; submission 580990, Wind_Rain_Tower Aug 2026 Official leaderboard
10 amanupg2 58.4% zero-shot; submission 581251, amanupg2 Aug 2026 Official leaderboard
11 StarAtNyte 57.7% submission 580781, StarAtNyte Aug 2026 Official leaderboard
12 Orpheon 57.6% zero-shot; submission 580466, Orpheon Aug 2026 Official leaderboard
13 AGI / AGI-KV-8 56.6% zero-shot; submission 580927, AGI Aug 2026 Official leaderboard
14 Lostagain 48.8% zero-shot; submission 581249, Lostagain Aug 2026 Official leaderboard
15 BoundedAccord / bounded-geometry-qwen-zero-shot 45.4% zero-shot; submission 579341, BoundedAccord Aug 2026 Official leaderboard
16 qianlan 45.1% zero-shot; submission 580716, qianlan Aug 2026 Official leaderboard
17 TamakiLab / KilometerVision Specialist Routing Best 42.7% zero-shot; submission 581246, TamakiLab Aug 2026 Official leaderboard
18 Openthu / Qwen3-VL-32B parent plus strict temporal graph specialists 40.9% 0-shot; submission 579930, Openthu Aug 2026 Official leaderboard
19 ZeroOne / Qwen3-VL-32B parent plus Qwen8 distance CLIP and strict temporal graph 40.9% 0-shot; submission 579937, ZeroOne Aug 2026 Official leaderboard
20 Math / Qwen3-VL-32B parent with strict temporal graph specialists 40.8% 0-shot; submission 579938, Math Aug 2026 Official leaderboard
21 LiouMIT / Qwen3-VL-32B three-expert consensus 38.8% 0-shot; submission 579904, LiouMIT Aug 2026 Official leaderboard
22 redde / redde-v7 33.1% zero-shot; submission 580670, redde Aug 2026 Official leaderboard
23 cbb 32.6% zero-shot; submission 581032, cbb Aug 2026 Official leaderboard
24 nexus_xxy / qwen 27.7% zero-shot; submission 581018, nexus_xxy Aug 2026 Official leaderboard
25 FZJ_IAS8 21.5% fine-tuned; submission 579743, FZJ_IAS8 Aug 2026 Official leaderboard
26 Host_37595_Team / zero-shot random baseline 19.6% zero-shot; submission 578763, Host_37595_Team Aug 2026 Official leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Vision

All 281 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%