Multimodal Agent Performance on MME-RealWorld Lite (Accuracy)
55.1AccuracyToolGate
Evaluation Results
| Method | Links | |
|---|---|---|
| ToolGateBackbone=235B-FP8, Training regime=Cross-domain2026.06 | 55.1 | |
| BaselineBackbone=235B-FP8, Training regime=None2026.06 | 52.6 | |
| ToolGateBackbone=30B, Training regime=Cross-domain2026.06 | 51.1 | |
| BaselineBackbone=30B, Training regime=None2026.06 | 50.8 | |
| ToolGateBackbone=30B, Training regime=In-domain2026.06 | 50.4 |