MultiChallenge

Multi-turn instruction following under realistic conversation constraints. Scale AI.

Published by

Scale AI SEAL

Category

Instruction following

Index weight

1.0

Models

29

Data as of

9 Sept 2026

Scale AI SEAL Leaderboards.

1. 1Muse Spark75.5%
2. 2Muse Spark 1.175.3%
3. 3Gemini 3.1 Pro71.4%
4. 4GPT-5.4 Pro69.2%
5. 5Gemini 3 Pro65.7%
6. 6GPT-5.163.4%
7. 7GPT-563.2%
8. 8o3-pro62.4%
9. 9Kimi K2.561.4%
10. 10Gemini 3.1 Flash Lite60.6%
11. 11GPT-5 mini59.0%
12. 12Claude Opus 4.559.0%
13. 13Claude Opus 458.6%
14. 14Claude Opus 4.157.2%
15. 15Claude Sonnet 457.1%

SearchAll providersAlibabaAnthropicGoogleMetaMoonshot AIOpenAIOpen weights29 of 29

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Muse SparkMeta	75.5%	–	65.4	–
2	Muse Spark 1.1Meta	75.3%	Muse Spark 1.1	64.9	–
3	Gemini 3.1 ProGoogle	71.4%	–	63.9	–
4	GPT-5.4 ProOpenAI	69.2%	–	64.2	–
5	Gemini 3 ProGoogle	65.7%	–	61.6	–
6	GPT-5.1OpenAI	63.4%	–	55.2	–
7	GPT-5OpenAI	63.2%	–	–	–
8	o3-proOpenAI	62.4%	–	54.4	–
9	Kimi K2.5Moonshot AIopen	61.4%	kimi-k2.5	56.1	–
10	Gemini 3.1 Flash LiteGoogle	60.6%	–	54.3	–
11	GPT-5 miniOpenAI	59.0%	–	–	–
12	Claude Opus 4.5Anthropic	59.0%	–	60.6	–
13	Claude Opus 4Anthropic	58.6%	–	54.4	–
14	Claude Opus 4.1Anthropic	57.2%	–	58.1	–
15	Claude Sonnet 4Anthropic	57.1%	–	54.0	–
16	o3OpenAI	56.6%	–	52.9	–
17	Claude Opus 4.6Anthropic	56.0%	–	55.3	–
18	Kimi K2 ThinkingMoonshot AIopen	55.4%	–	52.8	–
19	Claude Sonnet 4.5Anthropic	55.3%	–	52.8	–
20	Gemini 2.5 ProGoogle	53.6%	–	54.1	–
21	Claude 3.7 SonnetAnthropic	51.6%	–	52.1	–
22	GPT 5.1 InstantOpenAI	51.2%	–	–	–
23	Claude Haiku 4.5Anthropic	50.5%	–	48.0	–
24	gpt-oss-120bOpenAIopen	45.3%	–	47.6	–
25	o4-miniOpenAI	44.9%	–	48.0	–
26	Qwen3 235B A22BAlibabaopen	41.2%	–	48.7	–
27	GPT-4.1OpenAI	39.4%	gpt-4.1	47.3	–
28	Claude Opus 4.6Anthropic	37.1%	–	54.1	–
29	Gemini 2.0 FlashGoogle	36.4%	–	42.3	–