Benchmark · AgentCompetitive

APEX-Agents

APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.

Updated 2026-11-17
Models tested
79
Top score
84.2
Gemini 4 Carbon
Median
44.3
min 1.1
Top-5 spread
σ 3.1
Competitive

Best score over time · one chart, every benchmark

APEX-AGENTS74 MODELS · FRONTIER RUNNING MAX0255075100SCORE ↑Dec 24Jun 25Dec 25May 26Nov 26RELEASE DATE →benchgecko.ai/benchmark/apex-agents · frontier
Frontier on APEX-Agents rose from 17.2 to 84.2 in 19 months · +67.0 points · latest leader Gemini 4 Carbon from Google DeepMind.
Pink dots = frontier records · 17 totalClick to open model page

79 models tested · sorted by score

#ModelScore
1Google DeepMind logoGemini 4 Carbon84.2
2Google DeepMind logoGemini 4 Argon82.4
3Anthropic logoClaude Fable 5.280.1
4Google DeepMind logoGemini 4 Hydrogen78.0
5Anthropic logoClaude Sonnet 5.575.5
6Anthropic logoClaude Opus 5.573.5
7OpenAI logoGPT-6.2 Astra72.2
8xAI logoGrok 4.770.0
9DeepSeek logoDeepSeek V4.1 Flash 101769.8
10Anthropic logoClaude Fable 5.168.6
11OpenAI logoGPT-6.1 Terra68.3
12Google DeepMind logoGemini 3.7 Flash67.8
13Anthropic logoClaude Opus 565.8
14moonshotai logoKimi K3.1 Pro65.5
15xAI logoGrok 4.665.3
16OpenAI logoGPT-6 Astra64.7
17Google DeepMind logoGemini 3.8 Flash64.3
18Anthropic logoClaude Fable 563.6
19Alibaba Qwen logoQwen3.8 Max (0902)63.3
20OpenAI logoGPT-6.1 Sol60.0
21OpenAI logoGPT-5.6 Terra58.2
22z-ai logoGLM 5.356.6
23xAI logoGrok 4.556.2
24Anthropic logoClaude Haiku 5.556.1
25OpenAI logoGPT-5.555.1
26Anthropic logoClaude Sonnet 554.5
27OpenAI logoGPT-6 Sol54.3
28z-ai logoGLM 5.3 Flash52.8
29OpenAI logoGPT-5.452.4
30Mistral AI logoMistral Large 451.6
31OpenAI logoGPT-5.6 Sol51.4
32Anthropic logoClaude Opus 4.749.2
33Anthropic logoClaude Opus 4.848.9
34moonshotai logoKimi K347.8
35Alibaba Qwen logoQwen3.8 27B47.5
36DeepSeek logoDeepSeek V4 Pro 081347.3
37Google DeepMind logoGemini 3.6 Flash46.9
38Anthropic logoClaude Opus 4.646.3
39z-ai logoGLM 5.245.2
40OpenAI logoGPT-6 Luna44.3
41OpenAI logoGPT-5.6 Luna43.0
42Anthropic logoClaude Sonnet 4.643.0
43z-ai logoGLM 5.140.9
44DeepSeek logoDeepSeek V4.1 Flash39.5
45minimax logoMiniMax M337.7
46moonshotai logoKimi K2.7 Code37.6
47Google DeepMind logoGemini 3.1 Pro Preview35.1
48OpenAI logoGPT-5.234.3
49Meta logoMuse Spark Open33.4
50OpenAI logoGPT-5.3-Codex31.7
51Google DeepMind logoGemini 3.5 Flash Lite29.3
52Google DeepMind logoGemini 3.5 Flash27.5
53Alibaba Qwen logoQwen3.5 397B A17B24.9
54OpenAI logoGPT-5.4 Mini24.6
55Google DeepMind logoGemini 3 Flash Preview24.0
56NVIDIA logoNemotron 3 Ultra22.7
57DeepSeek logoDeepSeek V3.221.3
58Google DeepMind logoGemini 3 Pro18.4
59Anthropic logoClaude Opus 4.518.4
60OpenAI logoGPT-518.3
61OpenAI logoGPT-5.117.5
62OpenAI logoo317.2
63z-ai logoGLM 517.2
64OpenAI logoGPT-5.4 Nano16.9
65xAI logoGrok 415.2
66moonshotai logoKimi K2.514.4
67Alibaba Qwen logoQwen 3.5 Plus (hosted 397B-A17B)13.6
68Anthropic logoClaude Sonnet 49.3
69Anthropic logoClaude Haiku 4.58.9
70Google DeepMind logoGemini 2.5 Pro6.6
71minimax logoMiniMax M2.56.2
72OpenAI logogpt-oss-120b4.7
73moonshotai logoKimi K2 Thinking4.0
74z-ai logoGLM 4.73.1
75z-ai logoGLM 4.63.0
76xAI logoGrok 32.1
77Google DeepMind logoGemini 2.5 Flash1.8
78OpenAI logoo11.1
79OpenAI logoGPT-4o (2024-11-20)1.1

What it measures · how it is scored · sources · as of 2026-11-22

  • What it measures · Agent performance evaluation testing multi-step tool use, planning, and execution in realistic environments.
  • How it is scored · Scored as a percentage · higher is better · maximum 100
  • Released · 2026-01-21
  • Scores from · Epoch AI Benchmarking Hub (79 scores)
Details
Category
Agent
Max score
100
Models
79
Updated
2026-11-17

Same category · related evaluations