BareBones - WTP-Bench (Silhouette): leaderboard

BareBones (2026) tests zero-shot recognition from shape; its WTP-Bench set asks vision-language models to name each of 1,160 Pokemon from its sprite, scored as an exact species-and-form match of the top-1 answer. This board shows only the binarized silhouette, with no colour, texture or background. Accuracy in percent; higher is better.

Metric: Exact-match accuracy (%) naming the Pokemon from its pure-shape silhouette (no colour, texture or background); 1,160 Pokemon sprites, exact species-and-form match of the model's top-1 name (Master Ball tier), zero-shot, temperature 0 for API models and a 100-token output cap for open-weight models; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.553.1#261
2GPT-4.149.48#505
3Gemini 2.5 Flash30#463
4Claude Haiku 4.527.41#583
5Gemini 2.5 Pro23.02#276
6GPT-4o Mini8.53#1667
7Grok 4 Fast7.93#447
8Qwen 3 VL 4B3.1#1151
9LLaVA-OV-7B1.81#4475
10InternVL2.5-26B1.72#2156
11SmolVLM0.95
12LLaVA-v1.5-13B0.69#6926

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=barebones-wtp-bench-silhouette · How It Works · Data refreshed daily, snapshot 2026-11-21.