AgentWorldBench MCP

AgentWorldBench MCP domain - open-ended rubric evaluation (5 dimensions: Format, Factuality, Consistency, Realism, Quality) for language world models simulating MCP/tool-calling environments. Normalized to 0-100.

Category

General Agents

Max Score

100.0

Score Type

percent

Active

Yes

Model Scores

Scores for AgentWorldBench MCP

Model Score Date Verified
Qwen-AgentWorld-35B-A3B
65.00%
24.06.2026 Verified
Qwen-AgentWorld-397B-A17B
87.74%
24.06.2026 Unverified
Qwen3.5-35B-A3B
19.38%
24.06.2026 Unverified
Qwen3.5-397B-A17B
88.20%
24.06.2026 Unverified
Qwen3.6-Plus
2.31%
24.06.2026 Unverified
GPT-5.4
100.00%
24.06.2026 Unverified
Claude Opus 4.8
54.93
24.06.2026 Unverified
Claude Opus 4.6
98.68%
24.06.2026 Unverified
Gemini 3.1 Pro
27.29%
24.06.2026 Unverified
Claude Sonnet 4.6
99.34%
24.06.2026 Unverified
DeepSeek-V4-Pro
54.98%
24.06.2026 Unverified
GLM-5.1
83.52%
24.06.2026 Unverified
Kimi K2.6
67.90%
24.06.2026 Unverified
MiniMax-M2.7
5.87%
24.06.2026 Unverified