TutorBench

Tutoring tasks for high-school and AP subjects. Scale AI.

Published by

Scale AI SEAL

Category

Instruction following

Index weight

0.5

Models

27

Data as of

9 Sept 2026

Scale AI SEAL Leaderboards.

1. 1Muse Spark68.5%
2. 2GPT-5.4 Pro56.6%
3. 3Gemini 2.5 Pro55.6%
4. 4GPT-555.3%
5. 5o3-pro54.6%
6. 6Kimi K2.554.6%
7. 7GPT-5.154.1%
8. 8Claude Opus 4.653.7%
9. 9Gemini 3 Pro53.7%
10. 10Claude Opus 4.653.5%
11. 11GPT-5.253.5%
12. 12Gemini 3.1 Pro53.0%
13. 13o352.8%
14. 14o352.1%
15. 15Gemini 3.1 Flash Lite51.5%

SearchAll providersAnthropicGoogleMetaMoonshot AIOpenAIOpen weights27 of 27

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Muse SparkMeta	68.5%	–	65.4	–
2	GPT-5.4 ProOpenAI	56.6%	–	64.2	–
3	Gemini 2.5 ProGoogle	55.6%	–	54.1	–
4	GPT-5OpenAI	55.3%	–	59.5	–
5	o3-proOpenAI	54.6%	–	54.1	–
6	Kimi K2.5Moonshot AIopen	54.6%	kimi-k2.5	56.1	–
7	GPT-5.1OpenAI	54.1%	–	55.2	–
8	Claude Opus 4.6Anthropic	53.7%	–	54.1	–
9	Gemini 3 ProGoogle	53.7%	–	61.6	–
10	Claude Opus 4.6Anthropic	53.5%	–	55.3	–
11	GPT-5.2OpenAI	53.5%	–	61.5	–
12	Gemini 3.1 ProGoogle	53.0%	–	63.9	–
13	o3OpenAI	52.8%	–	53.6	–
14	o3OpenAI	52.1%	–	52.9	–
15	Gemini 3.1 Flash LiteGoogle	51.5%	–	54.3	–
16	Claude Opus 4.5Anthropic	51.2%	–	60.6	–
17	Claude Opus 4.1Anthropic	50.8%	–	58.1	–
18	Claude Opus 4.5Anthropic	49.8%	–	58.0	–
19	Claude Opus 4Anthropic	49.7%	–	54.4	–
20	GPT 5.1 InstantOpenAI	49.1%	–	–	–
21	Claude Sonnet 4.5Anthropic	49.0%	–	52.8	–
22	Claude Opus 4.1 AnthropicAnthropic	47.4%	–	–	–
23	Claude 37 SonnetAnthropic	46.5%	–	–	–
24	Claude Sonnet 4.5Anthropic	45.7%	–	52.3	–
25	Claude Opus 4Anthropic	45.5%	–	50.7	–
26	Llama 4 MaverickMetaopen	40.2%	–	42.6	–
27	GPT-4oOpenAI	36.1%	gpt-4o	42.3	–