GeoBrowse - Level 1 State (GATE Agent): leaderboard
GeoBrowse (2026) Level 1: models compose weak visual cues in a street scene into a location. This board scores the state-level targets with the model inside the GATE agent workflow of the benchmark with image and web tools; accuracy in percent from an LLM judge, mean of three runs, higher is better.
Metric: Accuracy (%) on the state-level targets of GeoBrowse Level 1 (199 street-scene instances whose location must be composed from weak, fragmented visual cues), judged correct by an LLM judge with the Humanity's Last Exam judging prompt, mean of 3 runs, under the benchmark's GATE agentic workflow: a ReAct loop with five image tools (crop, super-resolution and others) and four knowledge tools (web text and image search, page visits, code) under a fixed protocol, tool suite and call budget; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 42.6 | #145 |
| 2 | Claude Opus 4.5 | 32.7 | #160 |
| 3 | Gemini 2.5 Pro | 23.6 | #276 |
| 4 | GPT-5 | 23.6 | #178 |
| 5 | GPT-4o | 20 | #751 |
| 6 | GPT-4.1 | 20 | #505 |
| 7 | Qwen 3 VL 32B | 12.7 | #622 |
| 8 | Llama 3.2 90B | 10.9 | #1755 |
| 9 | Qwen 2.5 VL 7B | 9.1 | #1862 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=geobrowse-level-1-state-gate-agent · How It Works · Data refreshed daily, snapshot 2026-11-21.