livebench reasoning connections 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	DeepSeek V4 Pro 0813 DeepSeek	100%	100.0	69% confidence 69 percent, Medium
2	DeepSeek V4.1 Flash DeepSeek	100%	100.0	69% confidence 69 percent, Medium
3	Gemini 3.1 Pro Preview Google	100%	100.0	100% confidence 100 percent, Full
4	Gemini 3.5 Flash Google	100%	100.0	100% confidence 100 percent, Full
5	Gemini 3.6 Flash Google	100%	100.0	100% confidence 100 percent, Full
6	Gemini 3.7 Flash Google	100%	100.0	100% confidence 100 percent, Full
7	Gemini 3.8 Flash Google	100%	100.0	100% confidence 100 percent, Full
8	Muse Spark 1.2 Meta	100%	100.0	53% confidence 53 percent, Medium
9	Kimi K3 Moonshot AI	100%	100.0	100% confidence 100 percent, Full
10	GPT-5.4 OpenAI	100%	100.0	100% confidence 100 percent, Full
11	GPT-5.5 OpenAI	100%	100.0	100% confidence 100 percent, Full
12	GPT-5.6 Sol OpenAI	100%	100.0	100% confidence 100 percent, Full
13	GPT-6 Astra OpenAI	100%	100.0	100% confidence 100 percent, Full
14	GPT-6 Luna OpenAI	100%	100.0	100% confidence 100 percent, Full
15	GPT-6 Sol OpenAI	100%	100.0	100% confidence 100 percent, Full
16	GPT-6.1 Sol OpenAI	100%	100.0	100% confidence 100 percent, Full
17	Grok 4.6 xAI	100%	100.0	100% confidence 100 percent, Full
18	Claude Fable 5 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
19	Claude Fable 5.1 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
20	Claude Opus 4.5 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
21	Claude Opus 4.6 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
22	Claude Opus 4.8 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
23	Claude Opus 5 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
24	Claude Opus 5.5 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
25	Claude Sonnet 4.6 Anthropic	99.3%	99.3	100% confidence 100 percent, Full
26	DeepSeek V4 Flash Vision Exp DeepSeek	99.3%	99.3	21% confidence 21 percent, Low
27	GPT-5.2 OpenAI	99%	99.0	100% confidence 100 percent, Full
28	GPT-5.4 mini OpenAI	99.0%	99.0	100% confidence 100 percent, Full
29	Claude Sonnet 5.5 Anthropic	98.5%	98.5	93% confidence 93 percent, High
30	DeepSeek V4 Pro DeepSeek	98%	98.0	85% confidence 85 percent, High
31	Muse Spark 1.1 Meta	98%	98.0	61% confidence 61 percent, Medium
32	Kimi K2.7 Code Moonshot AI	98%	98.0	48% confidence 48 percent, Low
33	Grok 4.5 xAI	98%	98.0	100% confidence 100 percent, Full
34	Grok 4.7 xAI	98%	98.0	100% confidence 100 percent, Full
35	DeepSeek V4 Flash 0731 DeepSeek	97.3%	97.3	69% confidence 69 percent, Medium
36	Qwen3.7 Max Alibaba / Qwen	96.5%	96.5	53% confidence 53 percent, Medium
37	GPT-5.6 Luna OpenAI	96.5%	96.5	100% confidence 100 percent, Full
38	MiniMax-M3 minimax	96.3%	96.3	100% confidence 100 percent, Full
39	Claude Haiku 5.5 Anthropic	96%	96.0	100% confidence 100 percent, Full
40	Claude Sonnet 5 Anthropic	95.5%	95.5	93% confidence 93 percent, High
41	GLM-5.3 Z.ai	95.3%	95.3	93% confidence 93 percent, High
42	GPT-5.2 Codex OpenAI	95%	95.0	43% confidence 43 percent, Low
43	Qwen3.8 Max Alibaba / Qwen	94.5%	94.5	85% confidence 85 percent, High
44	GPT-5.4 nano OpenAI	94.4%	94.4	100% confidence 100 percent, Full
45	Muse Spark 1.3 Meta	94%	94.0	93% confidence 93 percent, High
46	Grok 4.3 xAI	94%	94.0	80% confidence 80 percent, High
47	GLM-5.2 Z.ai	94%	94.0	100% confidence 100 percent, Full
48	GLM-5.3-Flash Z.ai	93.7%	93.7	100% confidence 100 percent, Full
49	GPT-5.6 Terra OpenAI	93%	93.0	100% confidence 100 percent, Full
50	Claude Opus 4.7 Anthropic	92.3%	92.3	100% confidence 100 percent, Full
51	Gemini 3.5 Flash Lite Google	92%	92.0	100% confidence 100 percent, Full
52	Inkling thinkingmachines	91.2%	91.2	100% confidence 100 percent, Full
53	Qwen3.6 Plus Alibaba / Qwen	90.5%	90.5	80% confidence 80 percent, High
54	Qwen3.8 27B Alibaba / Qwen	90.5%	90.5	69% confidence 69 percent, Medium
55	Qwen3.8 Flash Next Alibaba / Qwen	90.2%	90.2	21% confidence 21 percent, Low
56	DeepSeek V4 Flash DeepSeek	89.5%	89.5	53% confidence 53 percent, Medium
57	Kimi K2.6 Moonshot AI	89.3%	89.3	85% confidence 85 percent, High
58	Nemotron 3 Ultra 550B A55B nvidia	86.7%	86.7	100% confidence 100 percent, Full
59	Grok Build 0.1 xAI	84%	84.0	16% confidence 16 percent, Low
60	Qwen3.6 27B Alibaba / Qwen	77.2%	77.2	53% confidence 53 percent, Medium
61	Mistral Large 4 mistral	48%	48.0	32% confidence 32 percent, Low

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.