Artificial Analysis · Checked 2026-10-02

Reasoning Ranking

Humanity's Last Exam as evaluated by Artificial Analysis; the source configuration is preserved.

Method Humanity's Last Exam as evaluated by Artificial Analysis; the source configuration is preserved.

Original scores from one independent snapshot. Each model keeps one evaluation configuration across the task lenses; rounded ties share a rank.

Rank	Model	Provider	Reasoning (%)	State	Coding (%)	Research (Elo)	Multimodal (%)	Agentic (%)
1	Claude Opus 5.5 max · adaptive reasoning · default fallback	Anthropic	61%	Measured ↗	60%	1846	26%	70%
2	Claude Fable 5.1 max · adaptive reasoning · default fallback	Anthropic	59%	Measured ↗	52%	1735	26%	59%
3	Gemini 4 Argon high · restricted access	Google	57%	Measured ↗	57%	1611	22%	78%
4	Claude Sonnet 5.5 max · adaptive reasoning · default fallback	Anthropic	55%	Measured ↗	64%	1844	26%	71%
4	GPT-6 Astra max	OpenAI	55%	Measured ↗	59%	1542	31%	68%
6	GPT-6.1 Sol max	OpenAI	53%	Measured ↗	56%	1575	31%	65%
7	Gemini 3.8 Flash high	Google	48%	Measured ↗	20%	1412	21%	60%
7	GPT-6 Sol max	OpenAI	48%	Measured ↗	44%	1505	25%	62%
9	Kimi K3 max	Moonshot AI	47%	Measured ↗	13%	1540	22%	58%
10	Grok 4.6 high	xAI	43%	Measured ↗	21%	1609	17%	67%
10	Grok 4.7 xhigh	xAI	43%	Measured ↗	26%	1695	20%	66%
10	Qwen3.8-Max 0803 release · reasoning enabled	Alibaba Cloud	43%	Measured ↗	19%	1608	20%	49%
10	Qwen3.8-Max-0902 0902 API snapshot · reasoning enabled	Alibaba Cloud	43%	Measured ↗	39%	1672	23%	56%
14	GLM-5.3 max	Z.ai	42%	Measured ↗	42%	1656	11%	62%
15	GLM-5.3-Flash AA evaluated Flash configuration	Z.ai	40%	Measured ↗	33%	1645	15%	60%
16	DeepSeek-V4.1-Flash max	DeepSeek	39%	Measured ↗	27%	1600	13%	69%
16	GPT-6 Luna max	OpenAI	39%	Measured ↗	13%	1438	23%	53%
16	MiniMax M3 AA evaluated M3 configuration	MiniMax	39%	Measured ↗	2%	1246	10%	21%
19	Mistral Medium 3.5 AA evaluated Medium 3.5 configuration	Mistral	14%	Measured ↗	0%	747	3%	6%

Other catalog models

No score for this exact benchmark in the selected snapshot. Available historical results remain in each profile.

27 other models

Outside this benchmark snapshot

Qwen3.8-Flash-Next

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Sonnet 5

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Haiku 4.5

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Mythos 5.1

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Gemini 3.8 Flash Cyber

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

GPT-5.6 Sol

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

GPT-5.6 Terra

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

GPT-5.6 Luna

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Gemini 3.5 Flash

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Gemini 3.6 Flash

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Fable 5

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Opus 5

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Opus 4.8

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Opus 4.7

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

DeepSeek V4 Pro

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

DeepSeek V4 Flash

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

OpenAI GPT-5.4

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

OpenAI GPT-5.2 Thinking

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Claude Opus 4.6

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Gemini 3.1 Pro

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Gemini 3 Pro

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

DeepSeek-R1

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Gemini 3.7 Flash

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Qwen3.8-Flash

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

DeepSeek-V3

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Tencent Hunyuan-Large

—

No score is inferred or assigned a ranking position.

Outside this benchmark snapshot

Hunyuan-T1

—

No score is inferred or assigned a ranking position.

Ranking source

Artificial Analysis Intelligence Index v4.3.2

Original metrics and evaluation configurations remain authoritative. Checked 2026-10-02.

Ranking source

Artificial Analysis methodology

Original metrics and evaluation configurations remain authoritative. Checked 2026-10-02.

Missing evidence is not zero. Higher reasoning effort may cost more, and small margins in rounded scores need not be significant. Full methodology →