APEX-Agents-AA · Open Analysis

#	Model	Effort	Score	Cost / task	Harness
1	Gemini 3.5 Flash	high	47.1%	–	–
2	Qwen 3.8 Max	max	42.4%	–	–
3	Kimi K3	max	41.3%	–	–
4	GPT-5.6 Terra	max	38.9%	–	–
5	Step 5 Preview	–	38.0%	–	–
6	GPT-5.5	xhigh	37.7%	–	–
7	GPT-5.6 Luna	max	35.8%	–	–
8	GLM 5.2	max	33.7%	–	–
9	GPT-5.4	xhigh	33.3%	–	–
10	Opus 4.6	max	33.0%	–	–
11	Gemini 3.1 Pro	–	32.0%	–	–
12	Apodex 1.1	–	31.2%	–	–
13	Kimi K2.6	–	28.5%	–	–
14	GPT-5.4 Mini	xhigh	28.2%	–	–
15	Sonnet 4.6	max	28.0%	–	–
16	Gemini 3 Flash	–	27.7%	–	–
17	Ling 3.0 Flash Fin	–	27.4%	–	–
18	GPT-5.4 Nano	xhigh	24.9%	–	–
19	DeepSeek V4 Pro 0424	max	24.3%	–	–
20	Qwen 3.7 Plus	–	22.4%	–	–
21	Grok 4.3	high	17.0%	–	–
22	Qwen 3.5 397B-A17B	–	15.3%	–	–
23	Step 3.7 Flash	high	14.8%	–	–
24	DeepSeek V3.2	–	14.5%	–	–
25	GLM 5	–	14.5%	–	–
26	Grok 4.20	–	14.2%	–	–
27	Gemini 3.1 Flash-Lite	high	12.2%	–	–
28	Kimi K2.5	–	11.5%	–	–
29	MiniMax M2.7	–	10.6%	–	–
30	gpt-oss-120b	high	3.1%	–	–
31	MiMo V2.5 Pro	–	2.4%	–	–
32	Nemotron 3 Super 120B A12B (Reasoning)	–	1.8%	–	–
33	gpt-oss-20b	high	0.7%	–	–