Benchmark Almanac

VANTAGE-BENCH

GPT-6 Astra 75.3 leads since September 2026
Not in model cards

Models answer questions, describe events, and locate objects or actions in infrastructure-camera footage. Task-specific metrics measure answer accuracy, caption quality, localization, and tracking.

The benchmark includes 35,027 annotations. The server holds the answer labels and scores submitted predictions.

Made by
NVIDIA
Released
September 2026
Kind
Video understanding
Level
Everyday
Scored
Checked automatically
Answers
Short answer
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Video, Images → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
A host
In views
Vision
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026

In the authors’ words

As Vision-Language Models (VLMs) advance toward physical deployment, the focus has remained on action-oriented Embodied AI evaluated on subject-centric consumer video. This overlooks a pervasive class of Physical AI: Infrastructure AI, which relies on fixed cameras for open-loop insights like safety monitoring and operational logging. We introduce VANTAGE-Bench, a benchmark measuring this "Infrastructure AI Gap." It spans three operational domains (Logistics, Transportation, and Smart Spaces), unifies image and video evaluation across semantic, spatial, temporal, and spatio-temporal capabilities, and moves beyond multiple-choice to eight task formulations including dense captioning and spatio-temporal grounding. It adds a single-pass trajectory protocol for Single Object Tracking and, to our knowledge, the first such evaluation on fixed-camera infrastructure video, scored against specialist trackers. Annotation spans three regimes over 3,346 media assets: 3,342 video-task annotations, 4,281 image-grounding annotations, and 27,404 detection boxes. Evaluating 17 models zero-shot, we find the shortfall relative to consumer-centric benchmarks is concentrated, not general. Event verification, referring expressions, and temporal localization fall roughly 9 to 24 points at every model scale, while video question answering stays within 5.3 points of VideoMME and 2D spatial pointing shows no shortfall against BLINK. The temporal pillar is weakest in absolute terms: no system exceeds 55.7 mIoU on temporal localization or 37.3 SODA_c on dense video captioning. On tracking, frontier models come within roughly 5 points of specialist trackers over short horizons but separate as the horizon extends. Open-weight models lead 2D object localization outright, so neither scale nor proprietary access explains the pattern. Data, evaluation harness, and leaderboard: https://vantage-bench.org/

Abstract of VANTAGE-Bench: Evaluating the Infrastructure AI Gap in Vision-Language Models (arXiv:2609.09396, CC0)
18
Entries on its own leaderboard, 14 models we track

GPT-6 Astra leads with 75.3 among models released by September 2026 GPT-6 Astra leads with 75.3 among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 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. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-6 Astra ✓ 75.3
  2. 2 Gemini 3.6 Flash ✓ 69.5
  3. 3 GPT-5.6 Sol ✓ 68.0
  4. 4 Cosmos3-Super ✓ 63.6
  5. 5 Gemini 3.1 Pro ✓ 62.7
  6. 6 Cosmos3-Nano ✓ 60.7
  7. 7 Qwen3-VL-32B-Instruct ✓ 55.4
  8. 8 Cosmos-Reason2-8B ✓ 54.5
  9. 9 Gemini 3.5 Flash-Lite ✓ 53.3
  10. 10 Gemini 3.1 Flash-Lite ✓ 52.6
  11. 11 Qwen3-VL-8B-Instruct ✓ 50.1
  12. 12 Qwen3.5-27B ✓ 49.3
  13. 13 Qwen3.5-9B ✓ 48.3
  14. 14 Cosmos-Reason2-32B ✓ 47.0
  15. 15 Cosmos3-Edge ✓ 46.9
  16. 16 Cosmos-Reason2-2B ✓ 45.9
  17. 17 Qwen3-VL-2B-Instruct ✓ 41.9
  18. 18 Gemma 4 E2B ✓ 21.3

Source: Official VANTAGE-BENCH leaderboard

#EntryScoreSettingReleasedRun onSource
1 GPT-6 Astra ✓ 75.3 Sep 2026 Oct 2026 Official leaderboard
2 Gemini 3.6 Flash ✓ 69.5 Jul 2026 Jul 2026 Official leaderboard
3 GPT-5.6 Sol ✓ 68.0 Jul 2026 Jul 2026 Official leaderboard
4 Cosmos3-Super ✓ open weights, 64B 63.6 May 2026 Official leaderboard
5 Gemini 3.1 Pro ✓ 62.7 Feb 2026 May 2026 Official leaderboard
6 Cosmos3-Nano ✓ 60.7 May 2026 May 2026 Official leaderboard
7 Qwen3-VL-32B-Instruct ✓ 55.4 Oct 2025 May 2026 Official leaderboard
8 Cosmos-Reason2-8B ✓ open weights, 8B 54.5 May 2026 Official leaderboard
9 Gemini 3.5 Flash-Lite ✓ 53.3 Jul 2026 Jul 2026 Official leaderboard
10 Gemini 3.1 Flash-Lite ✓ 52.6 Mar 2026 May 2026 Official leaderboard
11 Qwen3-VL-8B-Instruct ✓ 50.1 Oct 2025 May 2026 Official leaderboard
12 Qwen3.5-27B ✓ 49.3 Feb 2026 May 2026 Official leaderboard
13 Qwen3.5-9B ✓ 48.3 Feb 2026 May 2026 Official leaderboard
14 Cosmos-Reason2-32B ✓ open weights, 32B 47.0 May 2026 Official leaderboard
15 Cosmos3-Edge ✓ 46.9 Jul 2026 Jul 2026 Official leaderboard
16 Cosmos-Reason2-2B ✓ open weights, 2B 45.9 Jul 2026 Official leaderboard
17 Qwen3-VL-2B-Instruct ✓ 41.9 Oct 2025 Jul 2026 Official leaderboard
18 Gemma 4 E2B ✓ 21.3 Mar 2026 Jul 2026 Official leaderboard
#ModelScoreSettingReleasedRun onSource
1 GPT-6 Astra ✓ 75.3 Sep 2026 Oct 2026 Official leaderboard
2 Gemini 3.6 Flash ✓ 69.5 Jul 2026 Jul 2026 Official leaderboard
3 GPT-5.6 Sol ✓ 68.0 Jul 2026 Jul 2026 Official leaderboard
4 Gemini 3.1 Pro ✓ 62.7 Feb 2026 May 2026 Official leaderboard
5 Cosmos3-Nano ✓ 60.7 May 2026 May 2026 Official leaderboard
6 Qwen3-VL-32B-Instruct ✓ 55.4 Oct 2025 May 2026 Official leaderboard
7 Gemini 3.5 Flash-Lite ✓ 53.3 Jul 2026 Jul 2026 Official leaderboard
8 Gemini 3.1 Flash-Lite ✓ 52.6 Mar 2026 May 2026 Official leaderboard
9 Qwen3-VL-8B-Instruct ✓ 50.1 Oct 2025 May 2026 Official leaderboard
10 Qwen3.5-27B ✓ 49.3 Feb 2026 May 2026 Official leaderboard
11 Qwen3.5-9B ✓ 48.3 Feb 2026 May 2026 Official leaderboard
12 Cosmos3-Edge ✓ 46.9 Jul 2026 Jul 2026 Official leaderboard
13 Qwen3-VL-2B-Instruct ✓ 41.9 Oct 2025 Jul 2026 Official leaderboard
14 Gemma 4 E2B ✓ 21.3 Mar 2026 Jul 2026 Official leaderboard

Rankings by part

We read 12 parts from the benchmark’s own leaderboard in 2 groups: dimension and task type. Each ranks models on one part of the test. GPT-6 Astra leads 11 of them, and Qwen3.5-27B leads the other one. Arrows show how far a model moved from its overall rank among the models we track.

Semantic

14 models we track have a score here.

  1. 1 GPT-6 Astra 85.1
  2. 2 Gemini 3.6 Flash 79.4
  3. 3 GPT-5.6 Sol 77.1
  4. 4 Gemini 3.1 Pro 70.0
  5. 5 Cosmos3-Nano 68.9
  6. 6 Gemini 3.5 Flash-Lite ↑1 68.0
  7. 7 Gemini 3.1 Flash-Lite ↑1 65.9
  8. 8 Qwen3-VL-32B-Instruct ↓2 65.7
  9. 9 Cosmos3-Edge ↑3 63.0
  10. 10 Qwen3-VL-8B-Instruct ↓1 62.9

2 models have held number one on the official leaderboard since its release in September 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind OpenAI
Own leaderboard Not tested since Sep 2026

Source: Official VANTAGE-BENCH leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard
20.0 40.0 60.0 80.0 2026 20.0 40.0 60.0 80.0 2026

Source: Official VANTAGE-BENCH leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Audio and video

All 695 in Audio and video
  • Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
  • LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
  • MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
  • MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
  • FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better. Phi-4-multimodal 4.00%
  • VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%

Data audit

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (18 entries).
  • Pass Models scored 14 models we track, plus 4 other entries on its leaderboard.
  • Pass Newest score The newest is from September 2026, for GPT-6 Astra.
  • Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 14 rows for models we track and 4 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check