HiL-Bench

Whether agents notice information gaps and ask clarifying questions. Scale AI.

Published by

Scale AI SEAL

Category

Agents & tools

Index weight

0.5

Models

17

Data as of

9 Sept 2026

Scale AI SEAL Leaderboards.

1. 1Claude Fable 5.161.5%
2. 2Claude Opus 557.0%
3. 3Claude Fable 556.3%
4. 4GLM-5.243.7%
5. 5Claude Opus 4.741.7%
6. 6Gemini 3.8 Flash41.5%
7. 7GPT-5.539.7%
8. 8Claude Opus 4.638.3%
9. 9Claude Opus 4.835.3%
10. 10Gemini 3.1 Pro35.3%
11. 11GPT-5.6 Sol32.3%
12. 12Gemini 3.5 Flash27.7%
13. 13Grok 4.2020.0%
14. 14Kimi K2.618.7%
15. 15GPT-5.49.7%

SearchAll providersAnthropicGoogleMiniMaxMoonshot AIOpenAIxAIZhipu AIOpen weights17 of 17

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Claude Fable 5.1Anthropic	61.5%	Claude Fable 5.1	70.0	–
2	Claude Opus 5Anthropic	57.0%	Claude Opus 5	70.0	–
3	Claude Fable 5Anthropic	56.3%	Claude Fable 5	70.4	–
4	GLM-5.2Zhipu AIopen	43.7%	GLM 5.2	57.4	–
5	Claude Opus 4.7Anthropic	41.7%	Claude Opus 4.7	65.5	–
6	Gemini 3.8 FlashGoogle	41.5%	Gemini 3.8 Flash	62.4	–
7	GPT-5.5OpenAI	39.7%	GPT-5.5	66.6	–
8	Claude Opus 4.6Anthropic	38.3%	Claude Opus 4.6	63.1	–
9	Claude Opus 4.8Anthropic	35.3%	Claude Opus 4.8	64.7	–
10	Gemini 3.1 ProGoogle	35.3%	Gemini 3.1 Pro	63.9	–
11	GPT-5.6 SolOpenAI	32.3%	GPT 5.6 Sol	65.3	–
12	Gemini 3.5 FlashGoogle	27.7%	Gemini 3.5 Flash	60.7	–
13	Grok 4.20xAI	20.0%	Grok-4.20	57.9	–
14	Kimi K2.6Moonshot AIopen	18.7%	Kimi-k2.6	60.6	–
15	GPT-5.4OpenAI	9.7%	GPT-5.4	63.4	–
16	MiniMax-M2.5MiniMaxopen	6.3%	Minimax-M2.5	53.5	–
17	GPT-5.3 CodexOpenAI	4.3%	–	62.4	–