benchgap
Agentic · tools

MCP-Tasks leaderboard

As of 2026-11-21, the highest measured score on MCP-Tasks is 74.2% by Qwen3.5 397B. 48 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Atria Dawn Preview84.2%estimated ± 6.0 pp, low confidence
2Qwen3.8 Max84.2%estimated ± 6.0 pp, low confidence
3Qwen3.7 Max83.1%estimated ± 6.4 pp, low confidence
4Ornith-1.5-397B82.4%estimated ± 6.0 pp, low confidence
5GPT-5.6 Sol80.4%estimated ± 7.0 pp, low confidence
6Muse Spark80.4%estimated ± 6.4 pp, low confidence
7Gemini 3.5 Flash79.7%estimated ± 7.0 pp, low confidence
8MiMo-V2.5-Pro79.7%estimated ± 3.6 pp, low confidence
9GPT-5.579.3%estimated ± 7.0 pp, low confidence
10dots3-note Preview79.1%estimated ± 6.0 pp, low confidence
11Ornith-1.0-9B79.0%estimated ± 6.4 pp, low confidence
12LFM2.5-2.6B78.6%estimated ± 6.4 pp, low confidence
13GPT-5.6 Luna78.3%estimated ± 7.0 pp, low confidence
14GPT-5.6 Terra78.2%estimated ± 7.0 pp, low confidence
15MiMo-V2.577.5%estimated ± 6.4 pp, low confidence
16DeepSeek V4 Pro77.5%estimated ± 7.0 pp, low confidence
17Ling 3.0 Flash76.7%estimated ± 0.6 pp, low confidence
18Qwen3.6-35B-A3B76.6%estimated ± 0.6 pp, low confidence
19Spark-X2.6-4B76.6%estimated ± 0.6 pp, low confidence
20Solar Pro 476.5%estimated ± 0.6 pp, low confidence
21Step 3.7 Flash76.3%estimated ± 7.0 pp, low confidence
22Solar Open 276.3%estimated ± 0.6 pp, low confidence
23GPT-5.4 mini76.2%estimated ± 0.6 pp, low confidence
24GPT-5.4 nano76.0%estimated ± 0.6 pp, low confidence
25Kimi K2.676.0%estimated ± 0.6 pp, low confidence
26Nemotron 3.5 Bolt 74B-A9B BF1676.0%estimated ± 0.6 pp, low confidence
27Ornith-1.5-9B75.8%estimated ± 0.6 pp, low confidence
28GLM-5.275.6%estimated ± 7.0 pp, low confidence
29Nemotron 3.5 Bolt 74B-A9B75.4%estimated ± 0.6 pp, low confidence
30DeepSeek V4 Flash 073175.4%estimated ± 7.0 pp, low confidence
31Nemotron 3 Ultra75.1%estimated ± 3.6 pp, low confidence
32Qwen3.7 Plus74.9%estimated ± 1.3 pp, low confidence
33GLM-5.174.4%estimated ± 3.6 pp, medium confidence
34Qwen3.5 397B74.2%measured
35Qwen3.6 Plus74.1%measured
36LLaDA2.2-flash73.8%estimated ± 0.6 pp, medium confidence
37GPT-5.473.8%estimated ± 6.4 pp, low confidence
38LongCat-Flash-Lite-Sparse73.5%estimated ± 0.6 pp, medium confidence
39Claude Opus 4.571.8%measured
40Gemini 3.1 Pro69.3%estimated ± 6.4 pp, low confidence
41MiMo-V2-Pro69.3%estimated ± 6.4 pp, low confidence
42LLaDA2.2-mini68.1%estimated ± 6.4 pp, low confidence
43MiniMax M2.766.8%estimated ± 3.6 pp, medium confidence
44GLM-5-Turbo65.8%estimated ± 6.4 pp, low confidence
45Pokee-Isaac 28B63.1%estimated ± 3.6 pp, medium confidence
46LFM3-1.7B62.7%estimated ± 0.6 pp, medium confidence
47GLM-5V-Turbo62.4%estimated ± 6.4 pp, low confidence
48GLM-560.8%measured
49Kimi K2.559.1%measured
50Gemini 3 Flash54.8%estimated ± 6.4 pp, low confidence
51Ornith-1.5-35B-A3B54.8%estimated ± 6.0 pp, low confidence
52Granite 4.2 30B51.3%estimated ± 3.6 pp, low confidence
53MiMo-V2-Omni48.7%estimated ± 6.4 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General