benchmarks.site

Public leaderboard · Agents, tool use & computer use

Toolathlon (Tool Decathlon)

Tests AI agents on long, multi-step tasks across many everyday software tools.

Leaderboard

Top 3 right now

  1. 1stGLM 5.3 Flash (max)78.4 ± 1.9
  2. 2ndKimi K3 (max)76.5 ± 1.9
  3. 3rdClaude Opus 4.8 (max)76.2 ± 3.4

What it measures

Long-horizon tasks that span many real applications and tools, checked by execution-based verification.

How to read the numbers

Pass@1 (%), with margin — higher is better.

Results

7 rows · higher is better (Pass@1 (%))CSV
1GLM 5.3 Flash (max)78.4 ± 1.9
2Kimi K3 (max)76.5 ± 1.9
3Claude Opus 4.8 (max)76.2 ± 3.4
4Muse Spark 1.2 (xhigh)75.9 ± 1.3
5DeepSeek V4 Pro 0813 (max)74.4 ± 1.2
6GPT-5.5 (xhigh)73.5 ± 1.2
7Gemini 3.5 Flash (high)67.3 ± 1.2

Note: as_of is the most recent entry date shown; page has no explicit update date.

Source: Toolathlon team · toolathlon.xyz · results as of 2026-08-30 · retrieved 2026-11-17

Markdown · More Agents, tool use & computer use