BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

MCP Atlas

Real-world tool use through the Model Context Protocol. Scale AI.

As of 25 Oct 2026, Muse Spark 1.1 leads MCP Atlas on BenchLeader with 88.1%, ahead of Claude Fable 5.1 at 87.2%, across 33 model configurations with a published result.

Published by
Scale AI SEAL
Category
Agents & tools
Index weight
1.0
Models
33
Data as of
25 Oct 2026
Newest result
17 Sept 2026

Scale AI SEAL Leaderboards.

What the test looks like

MCP Atlas asks a model to complete real tasks by calling tools through the Model Context Protocol, such as querying services and chaining results, in Scale AI's environment.

How it is scored

Percent of tasks completed correctly.

What to keep in mind

Tool availability and descriptions shape the outcome; it measures practical tool use rather than reasoning depth.

Top 15
  1. 1Muse Spark 1.188.1%
  2. 2Claude Fable 5.187.2%
  3. 3Claude Opus 5 (xhigh)85.8%
  4. 4Qwen3.8 2.4T A95B (xhigh)84.5%
  5. 5GLM 5.384.2%
  6. 6Gemini 3.5 Flash (high)83.6%
  7. 7Claude Fable 583.3%
  8. 8Kimi K3 (max)82.3%
  9. 9Muse Spark82.2%
  10. 10Claude Opus 4.8 (max)82.2%
  11. 11GPT-5.681.8%
  12. 12Inkling Small79.2%
  13. 13Claude Opus 4.7 (max)79.1%
  14. 14Gemini 3.1 Pro (high)78.2%
  15. 15GLM 5.277.8%
33 of 33
#
1Muse Spark 1.188.1%62.52026-07-09
2Claude Fable 5.187.2%66.42026-09-04
3Claude Opus 5xhigh85.8%67.42026-08-03
4Qwen3.8 2.4T A95Bxhigh84.5%–2026-09-17
5GLM 5.384.2%–2026-09-17
6Gemini 3.5 Flashhigh83.6%63.12026-05-19
7Claude Fable 583.3%66.52026-06-09
8Kimi K3max82.3%64.82026-07-20
9Muse Spark82.2%64.32026-04-08
10Claude Opus 4.8max82.2%63.32026-05-28
11GPT-5.681.8%–2026-07-15
12Inkling Small79.2%54.92026-07-30
13Claude Opus 4.7max79.1%62.42026-04-08
14Gemini 3.1 Prohigh78.2%60.32025-11-24
15GLM 5.277.8%51.82026-07-20
16Claude Opus 4.6max76.8%57.42025-12-18
17GLM 5.175.6%55.92025-12-17
18GPT-5.5xhigh75.3%66.12026-04-28
19GPT-5.4xhigh70.6%64.72025-11-30
20Gemini 3 Pro70.3%60.42025-12-18
21Claude Opus 4.5high69.8%59.22025-09-10
22Claude Sonnet 4.669.5%57.12025-11-06
23GPT-5.2xhigh67.6%60.72025-09-10
24Kimi K2.564.4%54.52025-09-10
25Nemotron 3 Ultrathinking63.1%–2026-09-17
26Gemini 3 Flash62.0%57.02025-09-18
27Claude Sonnet 4.5thinking59.5%52.62025-12-16
28GLM 4.758.1%51.92025-09-10
29Gemini 3.1 Flash Litehigh57.1%48.82026-01-08
30GPT-5.4 Minixhigh56.7%56.32025-09-10
31GPT-5.1high50.1%57.72025-09-10
32o3-pro44.5%53.42025-09-10
33Claude Haiku 4.540.2%45.32025-09-10

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “MCP Atlas leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/scale_mcp_atlas.

MCP Atlas: questions

What does MCP Atlas measure?
MCP Atlas asks a model to complete real tasks by calling tools through the Model Context Protocol, such as querying services and chaining results, in Scale AI's environment. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads MCP Atlas?
Muse Spark 1.1 leads MCP Atlas with 88.1% as of 25 Oct 2026, ahead of Claude Fable 5.1 at 87.2%.
How many models have MCP Atlas results?
33 model configurations have a MCP Atlas result on BenchLeader, all taken from Scale AI SEAL.
Who runs MCP Atlas and how often is it updated?
MCP Atlas is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
Does MCP Atlas count toward the BenchLeader Index?
Yes. MCP Atlas contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.