Riemann-Bench · Open Analysis

#	Model	Effort	Score	Cost / task	Harness
1	GPT-5.6 Sol	max	74.4%	–	–
2	GPT-6 Astra	max	72.0%	–	–
3	Opus 5.5	max	69.6%	–	–
4	Opus 5	max	68.0%	–	–
5	GPT-5.6 Terra	max	66.4%	–	–
6	Fable 5.1	max	65.6%	–	–
7	GPT-6 Sol	max	64.8%	–	–
8	Fable 5	max	60.0%	–	–
9	GPT-6 Luna	max	59.2%	–	–
10	GPT-5.5	xhigh	55.2%	–	–
11	Gemini 3.8 Flash	high	51.2%	–	–
12	Opus 4.8	max	47.2%	–	–
13	GPT-5.4	xhigh	41.6%	–	–
14	Gemini 3.7 Flash	high	39.2%	–	–
15	DeepSeek V4 Pro	max	38.4%	–	–
16	Grok 4.5	high	38.4%	–	–
17	GPT-5.2	xhigh	37.6%	–	–
18	Kimi K3	max	37.6%	–	–
19	Gemini 3.5 Flash	high	36.8%	–	–
20	Gemini 3.1 Pro	high	33.6%	–	–
21	Opus 4.7	max	32.8%	–	–
22	Gemini 3.6 Flash	high	30.4%	–	–
23	Muse Spark 1.3	xhigh	28.0%	–	–
24	Opus 4.6	max	27.2%	–	–
25	DeepSeek V4.1 Flash	max	24.0%	–	–
26	DeepSeek V4 Flash	high	23.2%	–	–
27	Muse Spark 1.2	xhigh	23.2%	–	–
28	Muse Spark 1.1	xhigh	20.8%	–	–
29	Inkling	xhigh	15.2%	–	–
30	Qwen 3.7 Max	max	15.2%	–	–
31	Qwen 3.8 Max	xhigh	15.2%	–	–
32	GLM 5.3	max	12.0%	–	–
33	Kimi K2.5	max	12.0%	–	–
34	Opus 4.5	high	11.2%	–	–
35	Gemini 3.5 Flash-Lite	high	11.2%	–	–
36	GLM 5.2	max	10.4%	–	–
37	GLM 5.3 Flash	max	8.8%	–	–
38	DeepSeek V3.2	high	8.0%	–	–
39	Kimi K2.6	max	7.2%	–	–