Reasoning Capability

Reasoning and thinking benchmarks: HLE, ARC-AGI-2, GPQA Diamond.

698 models

#	Model	Developer				Open Source
1	Claude Mythos Preview	Anthropic	64.7	—	94.6	Closed
2	GPT-5.4 Pro	OpenAI	58.7	83.3	94.4	Closed
3	Muse Spark	Meta AI	58.0	42.5	89.5	Closed
4	GPT-5.5 Pro	OpenAI	57.2	84.6	—	Closed
5	Opus 4.7	Anthropic	54.7	75.8	94.2	Closed
6	Kimi K2.6	Moonshot AI	54.0	—	90.5	Closed
7	Qwen3.7-Max-Preview	アリババ	53.5	—	92.4	Closed
8	Claude Opus 4.6	Anthropic	53.0	66.3	91.3	Closed
9	GLM 5.1	Zhipu AI	52.3	—	—	Closed
10	GPT-5.5	OpenAI	52.2	85.0	93.6	Closed
11	GPT-5.4	OpenAI	52.1	77.1	92.8	Closed
12	Gemini 3.1 Pro Preview	Google DeepMind	51.4	77.1	94.3	Closed
13	Kimi K2 Thinking	Moonshot AI	51.0	—	—	Closed
14	Qwen 3.6 Plus Preview	アリババ	50.6	—	90.4	Closed
15	GLM-5	Zhipu AI	50.4	4.9	—	Closed
16	Kimi K2.5	Moonshot AI	50.2	11.8	—	Closed
17	Qwen3.6-Max-Preview	アリババ	50.2	—	90.4	Closed
18	GPT-5.2 Pro	OpenAI	50.0	54.2	93.2	Closed
19	Qwen3-Max-Thinking	アリババ	49.8	—	—	Closed
20	Claude Sonnet 4.6	Anthropic	49.0	58.3	89.9	Closed
21	Qwen3.5-27B	アリババ	48.5	—	—	Closed
22	Gemini 3 Deep Think - 2620	Google DeepMind	48.4	84.6	—	Closed
23	Qwen3.5-397B-A17B	アリババ	48.3	—	88.4	Closed
24	DeepSeek-V4-Pro	DeepSeek	48.2	—	89.1	Closed
25	Gemini 3.0 Pro (Preview 11-2025)	Google DeepMind	45.8	45.1	91.0	Closed
26	GPT-5.2	OpenAI	45.5	54.2	92.4	Closed
27	DeepSeek-V4-Flash	DeepSeek	45.1	—	88.1	Closed
28	Grok 4 Heavy	xAI	44.4	—	—	Closed
29	Gemini 3.0 Flash	Google DeepMind	43.5	33.6	90.4	Closed
30	Opus 4.5	Anthropic	43.2	37.6	—	Closed

About Benchmarks

HLE

総合知能テスト — 人間レベルの推論能力を測定

ARC-AGI-2

抽象的推論ベンチマーク — 新規パターンの汎化能力を測定

GPQA Diamond

Graduate-Level Google-Proof Q&A — 大学院レベルの科学的推論能力を測定