FrontierMath tiers 1-3

Mathematics · fixed question set

The lower three difficulty tiers of FrontierMath v2.

How far to trust it

Same private construction as FrontierMath v1. Reasoning effort moves this score more than any other benchmark on the site, so a figure with no effort setting attached says very little.

Measured

Models scored

Spread between models

27.1

standard deviation, points

Measurement noise

2.6

published stderr

Weight

0.99

share of spread that is signal

The weight is (spread² − noise²) / spread². It is computed, never chosen. Most of this benchmark's spread is real, so it still separates models.

Every model scored

the median across configurations, so one heroic run cannot lead

Rank	Model	Lab	Score
1	GPT-6.1 Sol	OpenAI	93.7
2	GPT-6 Astra	OpenAI	93.7
3	Claude Opus 5.5	Anthropic	91.2
4	Claude Fable 5.1	Anthropic	90.2
5	GPT-6 Sol	OpenAI	89.8
6	GPT-5.6 Sol	OpenAI	89.1
7	Claude Sonnet 5.5	Anthropic	88.8
8	GPT-5.5 Pro	OpenAI	87.7
9	Claude Fable 5	Anthropic	87.0
10	Gemini 4 Argon	Google	86.0
11	GPT-5.6 Terra	OpenAI	86.0
12	Claude Opus 5	Anthropic	85.6
13	GPT-5.5	OpenAI	85.3
14	GPT-5.4 Pro	OpenAI	82.5
15	Muse Wonder	Meta	82.1
16	Claude Opus 4.8	Anthropic	80.0
17	GPT-6 Luna	OpenAI	79.0
18	GPT-5.4	OpenAI	78.6
19	Claude Haiku 5.5	Anthropic	75.1
20	Qwen3.8 Max	Alibaba	74.7
21	Muse Spark 1.3	Meta	74.2
22	GPT-5.2 Pro	OpenAI	74.0
23	Kimi K3	Kimi	72.2
24	Gemini 3.7 Flash	Google	71.6
25	Claude Opus 4.7	Anthropic	70.2
26	Muse Spark Open	Meta	69.8
27	GLM-5.3	Zhipu	68.8
28	Gemini 3.8 Flash	Google	68.4
29	GPT 5.2	OpenAI	67.4
30	DeepSeek V4.1 Flash	DeepSeek	67.4
31	Claude Opus 4.6	Anthropic	66.0
32	Grok 4.6	xAI	66.0
33	Claude Sonnet 5	Anthropic	65.6
34	Qwen3.8 Max (0902)	Alibaba	65.6
35	DeepSeek V4 Pro 0813	DeepSeek	64.6
36	Qwen3.7-Max	Alibaba	64.6
37	Gemini 3.5 Flash	Google	62.8
38	GPT-5.6 Luna	OpenAI	61.8
39	Gemini 3.1 Pro Preview	Google	59.6
40	Gemini 3.6 Flash	Google	59.0
41	Mistral Large 4	Mistral	58.6
42	DeepSeek V4 Flash 0731	DeepSeek	57.5
43	Grok 4.5	xAI	57.2
44	Kimi K2.6	Kimi	57.2
45	GLM-5.2	Zhipu	57.0
46	GLM-5.3-Flash	Zhipu	55.8
47	GPT-5 Pro	OpenAI	55.8
48	Kimi K2.7 Code	Kimi	54.0
49	Grok 4.7	xAI	53.0
50	Gemini 3 Flash Preview	Google	51.2
51	Inkling Small	Thinking Machines	46.3
52	DeepSeek v4	DeepSeek	45.3
53	Grok 4.20	xAI	44.9
54	Grok 4.3 Beta	xAI	42.8
55	GLM-5.2	Zhipu	42.5
56	GPT-5.6 Luna	OpenAI	39.6
57	Qwen3.6 Plus	Alibaba	38.2
58	GPT-5.4 mini	OpenAI	37.9
59	GPT 5	OpenAI	37.2
60	Qwen3.6 27B	Alibaba	35.1
61	Claude Opus 4.5	Anthropic	34.4
62	Qwen3.7 Plus	Alibaba	34.4
63	Qwen3.6 27B	Alibaba	34.0
64	Inkling	Thinking Machines	33.3
65	GPT-5.4 nano	OpenAI	32.6
66	Qwen3.6 Plus	Alibaba	32.3
67	Qwen3.5 397B-A17B	Alibaba	31.2
68	GLM-5.1	Zhipu	30.9
69	Qwen3.5 397B-A17B	Alibaba	29.5
70	GPT 5 mini	OpenAI	29.4
71	o3	OpenAI	27.5
72	o4-mini	OpenAI	27.0
73	GPT-5.5 Instant	OpenAI	26.3
74	Gemini 3.5 Flash-Lite	Google	26.0
75	Gemini 2.5 Pro	Google	24.6
76	Claude Sonnet 4.5	Anthropic	23.9
77	Gemini 3.1 Flash-Lite	Google	22.5
78	Qwen3.6 Flash	Alibaba	22.5
79	Qwen3.6 35B-A3B	Alibaba	20.4
80	Qwen3.7 Flash	Alibaba	19.3
81	Qwen3.7 Flash	Alibaba	19.3
82	Qwen3 Max	Alibaba	18.9
83	Qwen3.5 Flash	Alibaba	18.2
84	Qwen3.6 35B-A3B	Alibaba	17.5
85	GPT-5.4 mini	OpenAI	17.2
86	Qwen3.6 Flash	Alibaba	17.2
87	Claude Opus 4.1	Anthropic	12.6
88	GPT 5 nano	OpenAI	11.9
89	o1-2024-12-17	OpenAI	11.1
90	O3 Mini	OpenAI	11.0
91	Qwen3.5 Flash	Alibaba	9.5
92	GPT 4.1 mini	OpenAI	6.7
93	GPT 4.1	OpenAI	6.0
94	GPT-5.4 nano	OpenAI	4.6
95	GPT-4 Turbo	OpenAI	0.7
96	GPT-4o mini	OpenAI	0.7
97	GPT 4o	OpenAI	0.3
98	GPT-3.5 Turbo	OpenAI	0.0

The benchmark's own page