Terminal-Bench Science-AA · Open Analysis

#	Model	Effort	Score	Cost / task	Harness
1	GPT-6 Astra	max	63.3%	–	–
2	Opus 5.5	xhigh	61.9%	–	–
3	GPT-6.1 Sol	max	58.1%	–	–
4	Sonnet 5.5	max	53.3%	–	–
5	Fable 5.1	max	43.3%	–	–
6	GPT-6 Sol	max	30.0%	–	–
7	Opus 5	max	28.6%	–	–
8	GPT-5.6 Sol	max	22.4%	–	–
9	Haiku 5.5	max	20.0%	–	–
10	GPT-5.6 Terra	max	13.8%	–	–
11	Qwen 3.8 Max	max	11.9%	–	–
12	Muse Spark 1.3	max	11.0%	–	–
13	Gemini 3.8 Flash	high	10.0%	–	–
14	Grok 4.7	high	9.5%	–	–
15	GLM 5.3	max	9.5%	–	–
16	DeepSeek V4.1 Flash	max	9.0%	–	–
17	GPT-6 Luna	max	8.6%	–	–
18	Grok 4.6	high	6.2%	–	–
19	MiMo V2.6 Pro	–	5.7%	–	–
20	DeepSeek V4 Pro 0813	max	5.7%	–	–
21	GLM 5.3 Flash	max	4.8%	–	–
22	MiMo V2.6 Flash	–	4.3%	–	–
23	GPT-5.6 Luna	max	3.3%	–	–
24	Step 5 Preview	–	2.4%	–	–
25	Qwen 3.8 27B	medium	1.0%	–	–
26	MiniMax M3	–	0.5%	–	–
27	K2 Horizon 375B A23B	–	0.0%	–	–
28	Inkling Small	–	0.0%	–	–
29	Inkling	xhigh	0.0%	–	–
30	Gemini 3.5 Flash-Lite	high	0.0%	–	–
31	Muse Glimmer (High)	high	0.0%	–	–