livebench reasoning theory of mind 2026_06_25 Benchmark: Scores and Sources

Published result; benchmark version and evaluation conditions remain in the id and result note.

pillar: reasoning · weight 1 within pillar · unit: % (higher is better) · official board ↗

Row	Model	Result	Normalized (0–100)	Confidence
1	GPT-5.4 OpenAI	88.5%	88.5	100% confidence 100 percent, Full
2	GLM-5.3 Z.ai	88.5%	88.5	93% confidence 93 percent, High
3	Claude Haiku 5.5 Anthropic	86.5%	86.5	100% confidence 100 percent, Full
4	DeepSeek V4 Flash 0731 DeepSeek	86.5%	86.5	69% confidence 69 percent, Medium
5	GPT-5.6 Terra OpenAI	86.5%	86.5	100% confidence 100 percent, Full
6	GPT-6.1 Sol OpenAI	86.5%	86.5	100% confidence 100 percent, Full
7	Grok 4.6 xAI	86.5%	86.5	100% confidence 100 percent, Full
8	Claude Fable 5 Anthropic	84.6%	84.6	100% confidence 100 percent, Full
9	Claude Opus 5.5 Anthropic	84.6%	84.6	100% confidence 100 percent, Full
10	DeepSeek V4 Pro 0813 DeepSeek	84.6%	84.6	69% confidence 69 percent, Medium
11	Muse Spark 1.3 Meta	84.6%	84.6	93% confidence 93 percent, High
12	GPT-5.5 OpenAI	84.6%	84.6	100% confidence 100 percent, Full
13	GPT-5.6 Sol OpenAI	84.6%	84.6	100% confidence 100 percent, Full
14	GPT-6 Astra OpenAI	84.6%	84.6	100% confidence 100 percent, Full
15	GPT-6 Sol OpenAI	84.6%	84.6	100% confidence 100 percent, Full
16	Grok 4.7 xAI	84.6%	84.6	100% confidence 100 percent, Full
17	Claude Opus 4.6 Anthropic	82.7%	82.7	100% confidence 100 percent, Full
18	Gemini 3.7 Flash Google	82.7%	82.7	100% confidence 100 percent, Full
19	Kimi K3 Moonshot AI	82.7%	82.7	100% confidence 100 percent, Full
20	Grok 4.5 xAI	82.7%	82.7	100% confidence 100 percent, Full
21	Qwen3.8 Flash Next Alibaba / Qwen	80.8%	80.8	21% confidence 21 percent, Low
22	Claude Fable 5.1 Anthropic	80.8%	80.8	100% confidence 100 percent, Full
23	Claude Opus 4.7 Anthropic	80.8%	80.8	100% confidence 100 percent, Full
24	Claude Opus 4.8 Anthropic	80.8%	80.8	100% confidence 100 percent, Full
25	Claude Sonnet 5 Anthropic	80.8%	80.8	93% confidence 93 percent, High
26	DeepSeek V4 Pro DeepSeek	80.8%	80.8	85% confidence 85 percent, High
27	DeepSeek V4.1 Flash DeepSeek	80.8%	80.8	69% confidence 69 percent, Medium
28	Gemini 3.1 Pro Preview Google	80.8%	80.8	100% confidence 100 percent, Full
29	Gemini 3.5 Flash Google	80.8%	80.8	100% confidence 100 percent, Full
30	Muse Spark 1.2 Meta	80.8%	80.8	53% confidence 53 percent, Medium
31	Mistral Large 4 mistral	80.8%	80.8	32% confidence 32 percent, Low
32	GPT-5.4 nano OpenAI	80.8%	80.8	100% confidence 100 percent, Full
33	Qwen3.7 Max Alibaba / Qwen	78.8%	78.8	53% confidence 53 percent, Medium
34	Qwen3.8 Max Alibaba / Qwen	78.8%	78.8	85% confidence 85 percent, High
35	Claude Opus 4.5 Anthropic	78.8%	78.8	100% confidence 100 percent, Full
36	Claude Opus 5 Anthropic	78.8%	78.8	100% confidence 100 percent, Full
37	DeepSeek V4 Flash Vision Exp DeepSeek	78.8%	78.8	21% confidence 21 percent, Low
38	Gemini 3.6 Flash Google	78.8%	78.8	100% confidence 100 percent, Full
39	GPT-5.2 OpenAI	78.8%	78.8	100% confidence 100 percent, Full
40	GPT-5.2 Codex OpenAI	78.8%	78.8	43% confidence 43 percent, Low
41	Gemini 3.8 Flash Google	76.9%	76.9	100% confidence 100 percent, Full
42	Muse Spark 1.1 Meta	76.9%	76.9	61% confidence 61 percent, Medium
43	MiniMax-M3 minimax	76.9%	76.9	100% confidence 100 percent, Full
44	GPT-5.4 mini OpenAI	76.9%	76.9	100% confidence 100 percent, Full
45	Kimi K2.6 Moonshot AI	75%	75.0	85% confidence 85 percent, High
46	GLM-5.2 Z.ai	75%	75.0	100% confidence 100 percent, Full
47	Claude Sonnet 4.6 Anthropic	73.1%	73.1	100% confidence 100 percent, Full
48	DeepSeek V4 Flash DeepSeek	73.1%	73.1	53% confidence 53 percent, Medium
49	GPT-5.6 Luna OpenAI	73.1%	73.1	100% confidence 100 percent, Full
50	GPT-6 Luna OpenAI	73.1%	73.1	100% confidence 100 percent, Full
51	GLM-5.3-Flash Z.ai	73.1%	73.1	100% confidence 100 percent, Full
52	Claude Sonnet 5.5 Anthropic	69.2%	69.2	93% confidence 93 percent, High
53	Kimi K2.7 Code Moonshot AI	69.2%	69.2	48% confidence 48 percent, Low
54	Grok Build 0.1 xAI	69.2%	69.2	16% confidence 16 percent, Low
55	Qwen3.6 Plus Alibaba / Qwen	67.3%	67.3	80% confidence 80 percent, High
56	Qwen3.6 27B Alibaba / Qwen	65.4%	65.4	53% confidence 53 percent, Medium
57	Inkling thinkingmachines	65.4%	65.4	100% confidence 100 percent, Full
58	Nemotron 3 Ultra 550B A55B nvidia	61.5%	61.5	100% confidence 100 percent, Full
59	Grok 4.3 xAI	61.5%	61.5	80% confidence 80 percent, High
60	Qwen3.8 27B Alibaba / Qwen	59.6%	59.6	69% confidence 69 percent, Medium
61	Gemini 3.5 Flash Lite Google	50%	50.0	100% confidence 100 percent, Full

Results are as published by the source behind each value (hover or tap the number). Benchmark scores are shown individually for every source/variant (a model may have multiple rows), and vary by version, harness and date; the normalized column uses the method’s fixed 0–100 scales and feeds the reasoning pillar of the SI Score.