APEX-Agents

Long-horizon professional agent tasks (Mercor).

Published by

Mercordata via Epoch AI Benchmarking Hub

Category

Agents & tools

Index weight

0.5

Models

63

Data as of

9 Sept 2026

CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.

1. 1Claude Fable 5.147.4%
2. 2GPT-6 Astra46.7%
3. 3Claude Fable 545.0%
4. 4Claude Fable 5.144.4%
5. 5Claude Opus 543.5%
6. 6Claude Opus 4.842.5%
7. 7Muse Spark 1.141.9%
8. 8Grok 4.641.2%
9. 9GPT-5.6 Sol40.0%
10. 10GPT-5.6 Sol39.9%
11. 11Kimi K339.3%
12. 12GPT-5.538.5%
13. 13GPT-5.538.5%
14. 14GPT-5.6 Sol37.7%
15. 15GPT-5.6 Sol37.7%

SearchAll providersAlibabaAnthropicGoogleMetaMiniMaxMoonshot AINVIDIAOpenAIxAIZhipu AIOpen weights63 of 63

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Claude Fable 5.1Anthropic	47.4%	–	70.0	–
2	GPT-6 AstraOpenAI	46.7%	GPT-6 Astra	69.9	–
3	Claude Fable 5Anthropic	45.0%	Fable 5	70.4	–
4	Claude Fable 5.1Anthropic	44.4%	–	69.7	–
5	Claude Opus 5Anthropic	43.5%	–	66.8	–
6	Claude Opus 4.8Anthropic	42.5%	–	59.4	–
7	Muse Spark 1.1Meta	41.9%	Muse Spark 1.1	64.9	–
8	Grok 4.6xAI	41.2%	Grok 4.6	63.6	–
9	GPT-5.6 SolOpenAI	40.0%	GPT 5.6 Sol (Max + Pro)	–	–
10	GPT-5.6 SolOpenAI	39.9%	–	65.5	–
11	Kimi K3Moonshot AIopen	39.3%	Kimi K3	66.5	–
12	GPT-5.5OpenAI	38.5%	GPT-5.5	66.6	–
13	GPT-5.5OpenAI	38.5%	–	63.7	–
14	GPT-5.6 SolOpenAI	37.7%	–	67.5	–
15	GPT-5.6 SolOpenAI	37.7%	GPT 5.6 Sol (xHigh)	65.3	–
16	GPT-5.4OpenAI	36.0%	–	61.3	–
17	GLM-5.2Zhipu AIopen	35.6%	GLM-5.2	57.4	–
18	GPT-5.4OpenAI	34.9%	–	63.4	–
19	GPT-5.2OpenAI	34.4%	–	57.3	–
20	Grok 4.5xAI	34.2%	Grok 4.5	63.4	–
21	Claude Opus 4.7Anthropic	33.9%	–	59.3	–
22	Gemini 3.1 ProGoogle	33.5%	–	63.9	–
23	Claude Sonnet 5Anthropic	32.5%	Sonnet 5	61.2	–
24	Claude Opus 4.6Anthropic	32.4%	Opus 4.6 (High)	63.1	–
25	Claude Opus 4.6Anthropic	32.1%	–	54.1	–
26	GPT-5.3 CodexOpenAI	31.8%	GPT 5.3 Codex	62.4	–
27	Gemini 3 ProGoogle	31.5%	–	61.6	–
28	GPT-5.2 CodexOpenAI	27.6%	GPT 5.2 Codex	60.7	–
29	Kimi K2.7 CodeMoonshot AIopen	27.6%	Kimi K2.7 Code	56.2	–
30	GPT-5.4 miniOpenAI	24.6%	–	54.8	–
31	Gemini 3 FlashGoogle	24.0%	–	56.6	–
32	Claude Sonnet 4.6Anthropic	23.7%	–	52.3	–
33	GPT-5.2OpenAI	23.0%	–	61.5	–
34	GPT-5.2OpenAI	23.0%	–	55.5	–
35	Claude Opus 4.5Anthropic	20.7%	–	58.0	–
36	GPT-5.1-CodexOpenAI	20.7%	GPT 5.1 Codex	57.1	–
37	GPT-5-CodexOpenAI	20.1%	–	–	–
38	Kimi K2.6Moonshot AIopen	18.9%	Kimi K2.6 (Thinking)	60.6	–
39	GPT-5OpenAI	18.3%	–	59.5	–
40	GPT-5OpenAI	18.3%	–	54.5	–
41	GPT-5.1OpenAI	17.5%	–	59.4	–
42	GPT-5.1OpenAI	17.5%	–	53.8	–
43	GLM-5Zhipu AIopen	17.2%	GLM 5 (Thinking)	59.0	–
44	o3OpenAI	17.2%	–	52.9	–
45	GPT-5.4 nanoOpenAI	16.9%	–	53.9	–
46	Grok 4xAI	15.2%	0709	57.2	–
47	Kimi K2.5Moonshot AIopen	14.4%	Kimi K2.5	56.1	–
48	Qwen3.5 PlusAlibaba	13.6%	Qwen 3.5 (Thinking)	–	–
49	Gemini 3.1 Flash LiteGoogle	13.0%	Gemini 3.1 Flash Lite	54.3	–
50	Grok 4.1xAI	12.8%	Grok 4.1	48.8	–
51	Nemotron 3 UltraNVIDIA	11.5%	Nemotron 3 Ultra (Thinking)	48.2	–
52	Claude Sonnet 4Anthropic	9.3%	–	49.5	–
53	Claude Haiku 4.5Anthropic	8.9%	–	47.9	–
54	GLM-4.7Zhipu AIopen	8.7%	GLM 4.7	54.9	–
55	Gemini 2.5 ProGoogle	6.6%	Gemini 2.5 Pro (On)	54.1	–
56	MiniMax-M2.5MiniMaxopen	6.2%	Minimax-2.5	53.5	–
57	gpt-oss-120bOpenAIopen	4.7%	GPT OSS 120B	47.6	–
58	Kimi K2 ThinkingMoonshot AIopen	4.1%	–	52.8	–
59	GLM-4.6Zhipu AIopen	4.0%	GLM 4.6	47.6	–
60	Grok 3xAI	2.1%	Grok 3	50.0	–

Show 3 more