BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

APEX-Agents

Long-horizon professional agent tasks (Mercor).

As of 24 Oct 2026, Gemini 4 Argon leads APEX-Agents on BenchLeader with 82.4%, ahead of Claude Sonnet 5.5 at 75.5%, across 53 model configurations with a published result.

Published by
Mercordata via Epoch AI Benchmarking Hub
Category
Agents & tools
Index weight
0.5
Models
53
Data as of
24 Oct 2026
Newest result
15 Oct 2026

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

What the test looks like

APEX-Agents sets long-horizon professional tasks, such as due diligence or a research literature review, that require many tool calls and hours of agent work, written and graded by domain experts at Mercor.

How it is scored

Percent of tasks judged successful by expert rubric, as published by Mercor.

What to keep in mind

Expert grading and long tasks mean small sample sizes and slow updates. Scores depend on the agent framework used in the published run.

Top 15
  1. 1Gemini 4 Argon82.4%
  2. 2Claude Sonnet 5.5 (max)75.5%
  3. 3Claude Opus 5.5 (max)73.5%
  4. 4Claude Fable 5.168.6%
  5. 5Gemini 3.7 Flash67.8%
  6. 6Claude Opus 5 (max)65.8%
  7. 7Grok 4.665.3%
  8. 8GPT-6 Astra64.7%
  9. 9Gemini 3.8 Flash64.3%
  10. 10Claude Fable 563.6%
  11. 11Qwen3.8 Max (max)63.3%
  12. 12GPT-6.1 Sol (max)60.0%
  13. 13Claude Fable 5.1 (high)59.7%
  14. 14Muse Spark 1.358.6%
  15. 15GPT-5.6 Terra (max)58.2%
53 of 53
#
1Gemini 4 Argon82.4%64.92026-09-30
2Claude Sonnet 5.5max75.5%69.32026-09-28
3Claude Opus 5.5max73.5%72.62026-09-22
4Claude Fable 5.168.6%66.42026-09-01
5Gemini 3.7 Flash67.8%53.42026-08-13
6Claude Opus 5max65.8%68.82026-07-24
7Grok 4.665.3%57.82026-08-12
8GPT-6 Astra64.7%68.02026-09-03
9Gemini 3.8 Flash64.3%55.72026-09-02
10Claude Fable 563.6%67.12026-06-09
11Qwen3.8 Maxmax63.3%63.32026-08-02
12GPT-6.1 Solmax60.0%69.92026-09-29
13Claude Fable 5.1high59.7%69.42026-09-01
14Muse Spark 1.358.6%59.82026-09-02
15GPT-5.6 Terramax58.2%64.42026-07-09
16GLM 5.356.6%–2026-08-14
17Grok 4.556.2%59.02026-07-08
18Claude Haiku 5.5max56.1%60.52026-10-07
19GPT-5.555.1%60.32026-04-23
20Grok 4.754.6%–2026-09-21
21Claude Sonnet 554.5%56.12026-06-30
22MiMo-V2.6-Pro54.4%64.02026-09-21
23GPT-6 Solmax54.3%66.42026-09-22
24GLM 5.3 Flash52.8%62.62026-08-20
25Claude Opus 5.5medium52.5%66.52026-09-22
26GPT-5.452.4%58.22026-03-05
27GPT-5.6 Sol Promax51.4%–2026-07-09
28Claude Opus 4.7max49.2%62.42026-04-16
29Claude Opus 4.8max48.9%63.32026-05-28
30Kimi K347.8%59.12026-07-16
31Muse Spark 1.3max47.6%67.12026-09-02
32Qwen3.8 27B47.5%54.62026-08-14
33DeepSeek V4 Pro47.3%53.52026-08-13
34Gemini 3.6 Flash46.9%49.52026-07-21
35Claude Opus 4.6max46.3%57.42026-02-05
36Muse Spark 1.245.4%–2026-08-05
37GLM 5.245.2%52.02026-06-16
38Claude Sonnet 5.5medium44.6%58.22026-09-28
39GPT-6 Lunamax44.3%60.02026-09-22
40Claude Sonnet 4.6high43.0%50.72026-02-17
41GPT-5.6 Luna43.0%48.02026-07-09
42GLM 5.140.9%55.92026-04-07
43DeepSeek V4.1 Flash39.5%–2026-09-09
44MiniMax M337.7%56.42026-06-01
45Kimi K2.7 Code37.6%55.82026-06-12
46Gemini 3.1 Pro35.1%60.62026-02-19
47NewMuse Spark Open 816B A42B33.4%–2026-10-15
48Muse Spark 1.131.8%62.52026-07-09
49Gemini 3.5 Flash Lite29.3%54.52026-07-21
50Gemini 3.5 Flash27.5%49.12026-05-19
51Qwen3.5 397B A17B24.9%51.52026-02-13
52Nemotron 3 Ultra22.7%44.12026-06-04
53gpt-oss-120b4.4%45.62025-08-05

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “APEX-Agents leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/apex_agents.

APEX-Agents: questions

What does APEX-Agents measure?
APEX-Agents sets long-horizon professional tasks, such as due diligence or a research literature review, that require many tool calls and hours of agent work, written and graded by domain experts at Mercor. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads APEX-Agents?
Gemini 4 Argon leads APEX-Agents with 82.4% as of 24 Oct 2026, ahead of Claude Sonnet 5.5 at 75.5%.
How many models have APEX-Agents results?
53 model configurations have a APEX-Agents result on BenchLeader, all taken from Mercor via Epoch AI Benchmarking Hub.
Who runs APEX-Agents and how often is it updated?
APEX-Agents is published by Mercor. BenchLeader re-reads the published results once a day and records the date each result was published.
Does APEX-Agents count toward the BenchLeader Index?
Yes. APEX-Agents contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.