PPBench · Open Analysis

#	Model	Effort	Score	Cost / task	Harness
1	Fable 5	max	97.6%	–	–
2	GPT-5.5	xhigh	83.3%	–	–
3	GPT-5.4	xhigh	70.2%	–	–
4	GPT-5.2	xhigh	56.0%	–	–
5	Opus 4.7	max	50.0%	–	–
6	Gemini 3.5 Flash	high	41.9%	–	–
7	Qwen 3.7 Max	max	40.0%	–	–
8	Opus 4.6	max	36.7%	–	–
9	Gemini 3.1 Pro	high	33.3%	–	–
10	Sonnet 4.6	max	26.7%	–	–
11	GLM 5.2	max	26.7%	–	–
12	GPT-5.2 Pro	xhigh	26.7%	–	–
13	Kimi K2.6	max	20.0%	–	–
14	Gemini 3 Pro	high	16.7%	–	–
15	Kimi K2.7 Code	max	16.7%	–	–
16	Qwen 3.7 Plus	high	16.7%	–	–
17	Qwen 3.6 Plus	max	10.0%	–	–
18	GPT-5.1	medium	7.7%	–	–
19	MiniMax M3	max	7.1%	–	–
20	Opus 4.5	high	6.7%	–	–
21	Gemini 3 Flash	high	6.7%	–	–
22	Grok 4.20	xhigh	6.7%	–	–
23	GPT-5	medium	6.0%	–	–
24	Kimi K2.5	max	6.0%	–	–
25	Grok 4.1 Fast	high	5.7%	–	–
26	DeepSeek V4 Pro	max	4.0%	–	–
27	Sonnet 4.5	high	3.3%	–	–
28	Grok 4.3	xhigh	3.3%	–	–
29	MiniMax M2.5	high	3.3%	–	–
30	Nemotron 3 Ultra	max	3.3%	–	–
31	o3	high	3.3%	–	–
32	DeepSeek V3.2	high	2.3%	–	–
33	Kimi K2	medium	1.3%	–	–
34	MiMo V2 Pro	–	1.0%	–	–
35	GLM 5	high	0.7%	–	–
36	MiniMax M2.7	high	0.7%	–	–
37	o1	high	0.7%	–	–
38	Qwen 3.5 397B-A17B	max	0.7%	–	–
39	Gemini 2.5 Pro	high	0.3%	–	–
40	Gemma 4 31B IT	high	0.3%	–	–
41	GLM 4.7	none	0.3%	–	–
42	gpt-oss-120b	high	0.3%	–	–
43	MiMo V2 Flash	high	0.3%	–	–
44	MiniMax M2.1	–	0.3%	–	–
45	Qwen 3 235B-A22B	xhigh	0.3%	–	–
46	Qwen 3	high	0.3%	–	–
47	Devstral 2512	–	0.0%	–	–
48	GPT-3.5	high	0.0%	–	–
49	GPT-4.1	high	0.0%	–	–
50	GPT-4o	high	0.0%	–	–
51	Mistral Large 2512	medium	0.0%	–	–
52	Mistral Small 2603	high	0.0%	–	–
53	Qwen 3 Coder	medium	0.0%	–	–