LiveBench Math: olympiadBenchmark 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 Fable 5.1 Anthropic	93.0%	93.0	100% confidence 100 percent, Full
2	Claude Fable 5 Anthropic	92.8%	92.8	100% confidence 100 percent, Full
3	Claude Opus 5 Anthropic	92.8%	92.8	100% confidence 100 percent, Full
4	Claude Opus 4.8 Anthropic	92.2%	92.2	100% confidence 100 percent, Full
5	Claude Opus 4.6 Anthropic	92.2%	92.2	100% confidence 100 percent, Full
6	GPT-6 Astra OpenAI	92.2%	92.2	100% confidence 100 percent, Full
7	Gemini 3.8 Flash Google	92.2%	92.2	100% confidence 100 percent, Full
8	Claude Opus 5.5 Anthropic	92.2%	92.2	100% confidence 100 percent, Full
9	Gemini 3.1 Pro Preview Google	92.1%	92.1	100% confidence 100 percent, Full
10	Gemini 3.5 Flash Google	91.9%	91.9	100% confidence 100 percent, Full
11	GPT-6.1 Sol OpenAI	91.9%	91.9	100% confidence 100 percent, Full
12	Gemini 3.7 Flash Google	91.8%	91.8	100% confidence 100 percent, Full
13	Muse Spark 1.3 Meta	91.7%	91.7	93% confidence 93 percent, High
14	Claude Sonnet 5.5 Anthropic	91.7%	91.7	93% confidence 93 percent, High
15	GPT-5.6 Sol OpenAI	91.7%	91.7	100% confidence 100 percent, Full
16	Grok 4.7 xAI	91.7%	91.7	100% confidence 100 percent, Full
17	Kimi K3 Moonshot AI	91.6%	91.6	100% confidence 100 percent, Full
18	GPT-5.6 Terra OpenAI	91.5%	91.5	100% confidence 100 percent, Full
19	GPT-5.4 OpenAI	91.5%	91.5	100% confidence 100 percent, Full
20	Claude Opus 4.7 Anthropic	91.4%	91.4	100% confidence 100 percent, Full
21	GPT-6 Sol OpenAI	91.4%	91.4	100% confidence 100 percent, Full
22	GPT-5.5 OpenAI	91.4%	91.4	100% confidence 100 percent, Full
23	DeepSeek V4 Pro 0813 DeepSeek	91.3%	91.3	69% confidence 69 percent, Medium
24	Grok 4.6 xAI	91.2%	91.2	100% confidence 100 percent, Full
25	Qwen3.8 Flash Next Alibaba / Qwen	91.2%	91.2	21% confidence 21 percent, Low
26	Qwen3.8 Max Alibaba / Qwen	91.2%	91.2	85% confidence 85 percent, High
27	Claude Sonnet 5 Anthropic	90.7%	90.7	93% confidence 93 percent, High
28	DeepSeek V4 Pro DeepSeek	90.6%	90.6	85% confidence 85 percent, High
29	Gemini 3.6 Flash Google	90.5%	90.5	100% confidence 100 percent, Full
30	GPT-6 Luna OpenAI	90.4%	90.4	100% confidence 100 percent, Full
31	Kimi K2.7 Code Moonshot AI	90.3%	90.3	48% confidence 48 percent, Low
32	DeepSeek V4 Flash Vision Exp DeepSeek	90.2%	90.2	21% confidence 21 percent, Low
33	Kimi K2.6 Moonshot AI	90.0%	90.0	85% confidence 85 percent, High
34	GLM-5.3-Flash Z.ai	89.8%	89.8	100% confidence 100 percent, Full
35	Claude Opus 4.5 Anthropic	89.5%	89.5	100% confidence 100 percent, Full
36	Mistral Large 4 Mistral AI	89.3%	89.3	64% confidence 64 percent, Medium
37	Grok 4.5 xAI	89.2%	89.2	100% confidence 100 percent, Full
38	DeepSeek V4 Flash 0731 DeepSeek	89.1%	89.1	69% confidence 69 percent, Medium
39	DeepSeek V4.1 Flash DeepSeek	89.1%	89.1	69% confidence 69 percent, Medium
40	GLM-5.2 Z.ai	89.0%	89.0	100% confidence 100 percent, Full
41	GPT-5.6 Luna OpenAI	88.6%	88.6	100% confidence 100 percent, Full
42	GPT-5.2 OpenAI	88.6%	88.6	100% confidence 100 percent, Full
43	GPT-5.4 nano OpenAI	88.0%	88.0	100% confidence 100 percent, Full
44	Claude Sonnet 4.6 Anthropic	87.9%	87.9	100% confidence 100 percent, Full
45	Muse Spark 1.2 Meta	87.8%	87.8	53% confidence 53 percent, Medium
46	Qwen3.8 27B Alibaba / Qwen	87.7%	87.7	69% confidence 69 percent, Medium
47	GLM-5.3 Z.ai	87.5%	87.5	93% confidence 93 percent, High
48	Muse Spark 1.1 Meta	87.5%	87.5	61% confidence 61 percent, Medium
49	GPT-5.2 Codex OpenAI	87.0%	87.0	43% confidence 43 percent, Low
50	Qwen3.7 Max Alibaba / Qwen	86.9%	86.9	53% confidence 53 percent, Medium
51	Grok Build 0.1 xAI	86.6%	86.6	16% confidence 16 percent, Low
52	DeepSeek V4 Flash DeepSeek	86.5%	86.5	53% confidence 53 percent, Medium
53	GPT-5.4 mini OpenAI	85.7%	85.7	100% confidence 100 percent, Full
54	MiniMax-M3 MiniMax	85.6%	85.6	100% confidence 100 percent, Full
55	Inkling Thinking Machines Lab	85.4%	85.4	100% confidence 100 percent, Full
56	Claude Haiku 5.5 Anthropic	83.7%	83.7	68% confidence 68 percent, Medium
57	Qwen3.6 Plus Alibaba / Qwen	82.8%	82.8	80% confidence 80 percent, High
58	Nemotron 3 Ultra 550B A55B NVIDIA	82.6%	82.6	100% confidence 100 percent, Full
59	Qwen3.6 27B Alibaba / Qwen	82.3%	82.3	53% confidence 53 percent, Medium
60	Grok 4.3 xAI	82.3%	82.3	80% confidence 80 percent, High
61	Gemini 3.5 Flash Lite Google	81.8%	81.8	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.

LiveBench Math: olympiadBenchmark scores and sources

What changed

Alerts on this device