livebench math amps hard 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 Opus 5 Anthropic	99.0%	99.0	100% confidence 100 percent, Full
2	Qwen3.8 Flash Next Alibaba / Qwen	99%	99.0	21% confidence 21 percent, Low
3	Claude Fable 5 Anthropic	99%	99.0	100% confidence 100 percent, Full
4	Claude Fable 5.1 Anthropic	99%	99.0	100% confidence 100 percent, Full
5	Claude Opus 4.5 Anthropic	99%	99.0	100% confidence 100 percent, Full
6	Claude Opus 5.5 Anthropic	99%	99.0	100% confidence 100 percent, Full
7	Gemini 3.8 Flash Google	99%	99.0	100% confidence 100 percent, Full
8	Muse Spark 1.3 Meta	99%	99.0	93% confidence 93 percent, High
9	Inkling thinkingmachines	99%	99.0	100% confidence 100 percent, Full
10	Grok 4.5 xAI	99%	99.0	100% confidence 100 percent, Full
11	GLM-5.3 Z.ai	99%	99.0	93% confidence 93 percent, High
12	Qwen3.7 Max Alibaba / Qwen	98%	98.0	53% confidence 53 percent, Medium
13	Qwen3.8 27B Alibaba / Qwen	98%	98.0	69% confidence 69 percent, Medium
14	Qwen3.8 Max Alibaba / Qwen	98%	98.0	85% confidence 85 percent, High
15	Claude Opus 4.7 Anthropic	98%	98.0	100% confidence 100 percent, Full
16	Claude Opus 4.8 Anthropic	98%	98.0	100% confidence 100 percent, Full
17	Claude Sonnet 5 Anthropic	98%	98.0	93% confidence 93 percent, High
18	Claude Sonnet 5.5 Anthropic	98%	98.0	93% confidence 93 percent, High
19	DeepSeek V4 Flash DeepSeek	98%	98.0	53% confidence 53 percent, Medium
20	DeepSeek V4 Pro DeepSeek	98%	98.0	85% confidence 85 percent, High
21	DeepSeek V4 Pro 0813 DeepSeek	98%	98.0	69% confidence 69 percent, Medium
22	DeepSeek V4.1 Flash DeepSeek	98%	98.0	69% confidence 69 percent, Medium
23	Gemini 3.1 Pro Preview Google	98%	98.0	100% confidence 100 percent, Full
24	Gemini 3.5 Flash Google	98%	98.0	100% confidence 100 percent, Full
25	Gemini 3.6 Flash Google	98%	98.0	100% confidence 100 percent, Full
26	Gemini 3.7 Flash Google	98%	98.0	100% confidence 100 percent, Full
27	Mistral Large 4 mistral	98%	98.0	32% confidence 32 percent, Low
28	GPT-5.2 Codex OpenAI	98%	98.0	43% confidence 43 percent, Low
29	GPT-5.4 OpenAI	98%	98.0	100% confidence 100 percent, Full
30	GPT-5.4 nano OpenAI	98%	98.0	100% confidence 100 percent, Full
31	GPT-5.6 Luna OpenAI	98%	98.0	100% confidence 100 percent, Full
32	GPT-5.6 Sol OpenAI	98%	98.0	100% confidence 100 percent, Full
33	GPT-5.6 Terra OpenAI	98%	98.0	100% confidence 100 percent, Full
34	GPT-6 Astra OpenAI	98%	98.0	100% confidence 100 percent, Full
35	GPT-6 Luna OpenAI	98%	98.0	100% confidence 100 percent, Full
36	GPT-6 Sol OpenAI	98%	98.0	100% confidence 100 percent, Full
37	GPT-6.1 Sol OpenAI	98%	98.0	100% confidence 100 percent, Full
38	Grok 4.7 xAI	98%	98.0	100% confidence 100 percent, Full
39	GLM-5.2 Z.ai	98%	98.0	100% confidence 100 percent, Full
40	Qwen3.6 Plus Alibaba / Qwen	97%	97.0	80% confidence 80 percent, High
41	Claude Haiku 5.5 Anthropic	97%	97.0	100% confidence 100 percent, Full
42	Claude Opus 4.6 Anthropic	97%	97.0	100% confidence 100 percent, Full
43	DeepSeek V4 Flash 0731 DeepSeek	97%	97.0	69% confidence 69 percent, Medium
44	Kimi K2.6 Moonshot AI	97%	97.0	85% confidence 85 percent, High
45	Kimi K2.7 Code Moonshot AI	97%	97.0	48% confidence 48 percent, Low
46	Kimi K3 Moonshot AI	97%	97.0	100% confidence 100 percent, Full
47	GPT-5.2 OpenAI	97%	97.0	100% confidence 100 percent, Full
48	GPT-5.4 mini OpenAI	97%	97.0	100% confidence 100 percent, Full
49	GPT-5.5 OpenAI	97%	97.0	100% confidence 100 percent, Full
50	Grok 4.3 xAI	97%	97.0	80% confidence 80 percent, High
51	Grok 4.6 xAI	97%	97.0	100% confidence 100 percent, Full
52	GLM-5.3-Flash Z.ai	97%	97.0	100% confidence 100 percent, Full
53	Nemotron 3 Ultra 550B A55B nvidia	96%	96.0	100% confidence 100 percent, Full
54	Gemini 3.5 Flash Lite Google	95%	95.0	100% confidence 100 percent, Full
55	Qwen3.6 27B Alibaba / Qwen	93%	93.0	53% confidence 53 percent, Medium
56	Muse Spark 1.2 Meta	89%	89.0	53% confidence 53 percent, Medium
57	DeepSeek V4 Flash Vision Exp DeepSeek	88%	88.0	21% confidence 21 percent, Low
58	Muse Spark 1.1 Meta	82%	82.0	61% confidence 61 percent, Medium
59	Claude Sonnet 4.6 Anthropic	76%	76.0	100% confidence 100 percent, Full
60	MiniMax-M3 minimax	75%	75.0	100% confidence 100 percent, Full
61	Grok Build 0.1 xAI	66%	66.0	16% confidence 16 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 math pillar of the SI Score.