Agentic · tools

Agents' Last Exam leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on Agents' Last Exam is 59.3% by GPT-6 Astra. 48 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	59.3%	measured
2	GPT-6.1 Sol	57.5%	estimated ± 6.3 pp, medium confidence
3	GPT-6 Sol	56.4%	measured
4	Qwen3.8 Max	52.4%	measured
5	Muse Spark 1.3	51.4%	estimated ± 6.3 pp, medium confidence
6	Qwen3.8-Flash-Next	51.2%	measured
7	Claude Fable 5.1	50.8%	estimated ± 6.3 pp, medium confidence
8	Claude Opus 5.5	50.8%	estimated ± 6.3 pp, medium confidence
9	Claude Sonnet 5.5	50.2%	estimated ± 6.3 pp, medium confidence
10	Claude 4.1 Opus	48.9%	estimated ± 8.7 pp, low confidence
11	Claude 4 Sonnet	48.9%	estimated ± 8.7 pp, low confidence
12	Claude Haiku 4.5	48.9%	estimated ± 8.7 pp, low confidence
13	Claude Sonnet 4.5	48.9%	estimated ± 8.7 pp, low confidence
14	Gemini 3 Flash	48.9%	estimated ± 8.7 pp, low confidence
15	Gemini 3 Pro	48.9%	estimated ± 8.7 pp, low confidence
16	GPT-5.1-Codex	48.9%	estimated ± 8.7 pp, low confidence
17	GPT-5.2	48.9%	estimated ± 8.7 pp, low confidence
18	GPT-5.2-Codex	48.9%	estimated ± 8.7 pp, low confidence
19	GPT-5.3 Codex	48.9%	estimated ± 8.7 pp, low confidence
20	GPT-5 (high)	48.9%	estimated ± 8.7 pp, low confidence
21	Kimi K2.5	48.9%	estimated ± 8.7 pp, low confidence
22	Qwen3.5 Plus	48.9%	estimated ± 8.7 pp, low confidence
23	GPT-6 Luna	45.7%	estimated ± 6.3 pp, medium confidence
24	Kimi K3	44.5%	estimated ± 6.3 pp, medium confidence
25	Gemini 3.8 Flash	42.9%	estimated ± 6.3 pp, medium confidence
26	Qwen3.8-27B	42.9%	measured
27	Claude Haiku 5.5	42.6%	estimated ± 6.3 pp, medium confidence
28	Grok 4.7	41.3%	estimated ± 6.3 pp, medium confidence
29	Gemini 4 Argon	39.5%	measured
30	Mistral Large 4	39.0%	estimated ± 6.3 pp, medium confidence
31	DeepSeek V4.1 Flash	31.8%	measured
32	MiMo-V2.6-Pro	31.6%	measured
33	Ling 3.1 Flash	30.3%	estimated ± 2.1 pp, high confidence
34	Fugu Cyber	30.3%	estimated ± 2.1 pp, high confidence
35	Atria Dawn Preview	30.3%	estimated ± 2.1 pp, high confidence
36	Gemini 3.8 Flash Cyber	30.3%	estimated ± 2.1 pp, high confidence
37	Step 5 Preview	29.5%	measured
38	GPT-5.6 Sol	28.8%	estimated ± 2.1 pp, high confidence
39	Inkling	28.5%	estimated ± 6.3 pp, medium confidence
40	GLM-5.3	28.5%	measured
41	MiMo-V2.6-Flash	27.6%	measured
42	Claude Mythos 5	27.0%	estimated ± 2.1 pp, high confidence
43	Gemini 3.7 Flash	26.3%	measured
44	GLM-5.3-Flash	26.3%	measured
45	DeepSeek V4 Pro 0813	25.7%	measured
46	Gemini 3.5 Flash Cyber	25.4%	estimated ± 2.1 pp, high confidence
47	Claude Mythos Preview	25.2%	estimated ± 2.1 pp, high confidence
48	DeepSeek V4 Flash 0731	25.2%	measured
49	GPT-5.5	24.2%	estimated ± 2.1 pp, high confidence
50	GPT-5.6 Terra	24.2%	estimated ± 2.1 pp, high confidence
51	GPT-5.4	24.0%	estimated ± 2.1 pp, high confidence
52	GPT-5.6 Luna	24.0%	estimated ± 2.1 pp, high confidence
53	Claude Opus 4.5	24.0%	estimated ± 2.1 pp, medium confidence
54	Claude Opus 4.6	24.0%	estimated ± 2.1 pp, medium confidence
55	Claude Opus 4.7 (Adaptive)	24.0%	estimated ± 2.1 pp, medium confidence
56	Claude Sonnet 4.6	24.0%	estimated ± 2.1 pp, medium confidence
57	GLM-5	24.0%	estimated ± 2.1 pp, medium confidence
58	GLM-5.1	24.0%	estimated ± 2.1 pp, medium confidence
59	Muse Spark	24.0%	estimated ± 2.1 pp, medium confidence
60	Muse Spark 1.1	24.0%	estimated ± 2.1 pp, medium confidence
61	Muse Glimmer 30B	23.0%	estimated ± 6.3 pp, low confidence
62	Hy4 preview	22.8%	measured
63	MiniMax M3	22.6%	estimated ± 6.3 pp, low confidence
64	Nemotron 3 Ultra	12.2%	estimated ± 6.3 pp, low confidence

Agents' Last Exam leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General