livebench math math comp 2026_06_25 Benchmark: Scores and Sources

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

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

Row	Model	Result	Normalized (0–100)	Confidence
1	Claude Haiku 5.5 Anthropic	98.0%	98.0	100% confidence 100 percent, Full
2	Claude Opus 4.7 Anthropic	98.0%	98.0	100% confidence 100 percent, Full
3	Claude Opus 4.8 Anthropic	98.0%	98.0	100% confidence 100 percent, Full
4	Qwen3.7 Max Alibaba / Qwen	97.1%	97.1	53% confidence 53 percent, Medium
5	Claude Fable 5.1 Anthropic	97.1%	97.1	100% confidence 100 percent, Full
6	Claude Opus 5.5 Anthropic	97.1%	97.1	100% confidence 100 percent, Full
7	Claude Sonnet 5.5 Anthropic	97.1%	97.1	93% confidence 93 percent, High
8	DeepSeek V4 Flash DeepSeek	97.1%	97.1	53% confidence 53 percent, Medium
9	DeepSeek V4 Flash Vision Exp DeepSeek	97.1%	97.1	21% confidence 21 percent, Low
10	DeepSeek V4 Pro 0813 DeepSeek	97.1%	97.1	69% confidence 69 percent, Medium
11	Muse Spark 1.2 Meta	97.1%	97.1	53% confidence 53 percent, Medium
12	Muse Spark 1.3 Meta	97.1%	97.1	93% confidence 93 percent, High
13	Mistral Large 4 mistral	97.1%	97.1	32% confidence 32 percent, Low
14	GPT-6 Astra OpenAI	97.1%	97.1	100% confidence 100 percent, Full
15	GPT-6 Sol OpenAI	97.1%	97.1	100% confidence 100 percent, Full
16	GPT-6.1 Sol OpenAI	97.1%	97.1	100% confidence 100 percent, Full
17	Inkling thinkingmachines	97.1%	97.1	100% confidence 100 percent, Full
18	Grok 4.5 xAI	97.1%	97.1	100% confidence 100 percent, Full
19	Grok 4.6 xAI	97.1%	97.1	100% confidence 100 percent, Full
20	Grok 4.7 xAI	97.1%	97.1	100% confidence 100 percent, Full
21	DeepSeek V4 Flash 0731 DeepSeek	96.1%	96.1	69% confidence 69 percent, Medium
22	DeepSeek V4 Pro DeepSeek	96.1%	96.1	85% confidence 85 percent, High
23	DeepSeek V4.1 Flash DeepSeek	96.1%	96.1	69% confidence 69 percent, Medium
24	Gemini 3.1 Pro Preview Google	96.1%	96.1	100% confidence 100 percent, Full
25	Gemini 3.7 Flash Google	96.1%	96.1	100% confidence 100 percent, Full
26	Muse Spark 1.1 Meta	96.1%	96.1	61% confidence 61 percent, Medium
27	Kimi K2.6 Moonshot AI	96.1%	96.1	85% confidence 85 percent, High
28	Nemotron 3 Ultra 550B A55B nvidia	96.1%	96.1	100% confidence 100 percent, Full
29	GPT-5.5 OpenAI	96.1%	96.1	100% confidence 100 percent, Full
30	GLM-5.2 Z.ai	96.1%	96.1	100% confidence 100 percent, Full
31	GLM-5.3 Z.ai	96.1%	96.1	93% confidence 93 percent, High
32	Qwen3.8 Max Alibaba / Qwen	95.1%	95.1	85% confidence 85 percent, High
33	Claude Fable 5 Anthropic	95.1%	95.1	100% confidence 100 percent, Full
34	Claude Opus 4.5 Anthropic	95.1%	95.1	100% confidence 100 percent, Full
35	Claude Opus 4.6 Anthropic	95.1%	95.1	100% confidence 100 percent, Full
36	Claude Sonnet 5 Anthropic	95.1%	95.1	93% confidence 93 percent, High
37	Gemini 3.5 Flash Google	95.1%	95.1	100% confidence 100 percent, Full
38	Gemini 3.8 Flash Google	95.1%	95.1	100% confidence 100 percent, Full
39	Kimi K3 Moonshot AI	95.1%	95.1	100% confidence 100 percent, Full
40	GPT-5.2 OpenAI	95.1%	95.1	100% confidence 100 percent, Full
41	GPT-5.2 Codex OpenAI	95.1%	95.1	43% confidence 43 percent, Low
42	GPT-5.4 nano OpenAI	95.1%	95.1	100% confidence 100 percent, Full
43	GPT-5.6 Sol OpenAI	95.1%	95.1	100% confidence 100 percent, Full
44	GPT-5.6 Terra OpenAI	95.1%	95.1	100% confidence 100 percent, Full
45	Grok 4.3 xAI	95.1%	95.1	80% confidence 80 percent, High
46	Grok Build 0.1 xAI	95.1%	95.1	16% confidence 16 percent, Low
47	Qwen3.8 27B Alibaba / Qwen	94.1%	94.1	69% confidence 69 percent, Medium
48	Qwen3.8 Flash Next Alibaba / Qwen	94.1%	94.1	21% confidence 21 percent, Low
49	Claude Opus 5 Anthropic	94.1%	94.1	100% confidence 100 percent, Full
50	Claude Sonnet 4.6 Anthropic	94.1%	94.1	100% confidence 100 percent, Full
51	Gemini 3.6 Flash Google	94.1%	94.1	100% confidence 100 percent, Full
52	Kimi K2.7 Code Moonshot AI	94.1%	94.1	48% confidence 48 percent, Low
53	GPT-5.4 OpenAI	94.1%	94.1	100% confidence 100 percent, Full
54	GPT-5.4 mini OpenAI	94.1%	94.1	100% confidence 100 percent, Full
55	GPT-6 Luna OpenAI	94.1%	94.1	100% confidence 100 percent, Full
56	Qwen3.6 Plus Alibaba / Qwen	93.1%	93.1	80% confidence 80 percent, High
57	GLM-5.3-Flash Z.ai	93.1%	93.1	100% confidence 100 percent, Full
58	Qwen3.6 27B Alibaba / Qwen	92.2%	92.2	53% confidence 53 percent, Medium
59	Gemini 3.5 Flash Lite Google	92.2%	92.2	100% confidence 100 percent, Full
60	GPT-5.6 Luna OpenAI	92.2%	92.2	100% confidence 100 percent, Full
61	MiniMax-M3 minimax	91.2%	91.2	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 math pillar of the SI Score.