Skip to content
Weights Not Walls
Menu

SpatialViz-Bench

Reported

No description yet: we only describe an eval set when we can do it correctly.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Link
https://arxiv.org/abs/2507.07610
Score shown
Epoch's column "Overall score", as a percentage, higher is better. The source gives no error bars.
Results
8 in all: 3 for open-weight models, 4 for closed ones, 1 where Epoch does not say. 2 linked to a model page here.

Best open-weight result: meta-llama/Llama-4-Scout-17B-16E-Instruct, 34.2%, number 4 of 8 results.

SpatialViz-Bench results
#ModelWeightsSettingScoreDateSource
1gemini-2.5-proGoogle DeepMindclosedPrompting: CoT44.7%not datedarxiv.org
2o1-2024-12-17_unknownOpenAIclosedeffort unknown; Prompting: CoT41.4%not datedarxiv.org
3gemini-2.5-flashGoogle DeepMindclosedPrompting: CoT36.9%not datedarxiv.org
4meta-llama/Llama-4-Scout-17B-16E-InstructEpoch: Llama-4-Scout-17B-16E-Instruct, Meta AIopenPrompting: CoT34.2%not datedarxiv.org
5claude-3-7-sonnet-20250219_unknownAnthropicclosedeffort unknown; Prompting: CoT33.9%not datedarxiv.org
6Qwen2.5-VL-72B-InstructAlibabano page matchopenPrompting: CoT33.3%not datedarxiv.org
7Qwen-VL-Maxnot statedPrompting: CoT32.0%not datedarxiv.org
8meta-llama/Llama-4-Maverick-17B-128E-InstructEpoch: Llama-4-Maverick-17B-128E-Instruct, Meta AIopenPrompting: CoT31.8%not datedarxiv.org

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method