General AI Assistant Tasks on GAIA (Task success rate)
71.5Task Success RatePIVOT
Evaluation Results
| Method | Links | |
|---|---|---|
| PIVOTBackbone=Opus 4.6, Feedback regime=w/ HITL2026.05 | 71.5 | |
| Self-CritiqueBackbone=Opus 4.62026.05 | 68.5 | |
| PIVOTBackbone=Opus 4.5, Feedback regime=w/ HITL2026.05 | 68.5 | |
| AgentDebugBackbone=Opus 4.62026.05 | 66.7 | |
| PIVOTBackbone=Opus 4.6, Feedback regime=w/o HITL2026.05 | 64.8 | |
| Self-CritiqueBackbone=Opus 4.52026.05 | 64.2 | |
| AgentDebugBackbone=Opus 4.52026.05 | 63.6 | |
| PIVOTBackbone=Opus 4.5, Feedback regime=w/o HITL2026.05 | 60.6 | |
| ReACTBackbone=Opus 4.52026.05 | 59.4 | |
| SE-AgentBackbone=Opus 4.52026.05 | 57.6 | |
| ReACTBackbone=Opus 4.62026.05 | 57 | |
| SE-AgentBackbone=Opus 4.62026.05 | 56.4 | |
| PIVOTBackbone=Sonnet 4.5, Feedback regime=w/ HITL2026.05 | 52.7 | |
| Self-CritiqueBackbone=Sonnet 4.52026.05 | 51.5 | |
| AgentDebugBackbone=Sonnet 4.52026.05 | 47.9 | |
| Self-CritiqueBackbone=Haiku 4.52026.05 | 44.8 | |
| PIVOTBackbone=Haiku 4.5, Feedback regime=w/ HITL2026.05 | 44.8 | |
| AgentDebugBackbone=Haiku 4.52026.05 | 41.8 | |
| PIVOTBackbone=Sonnet 4.5, Feedback regime=w/o HITL2026.05 | 41.8 | |
| PIVOTBackbone=Q3 235B, Feedback regime=w/ HITL2026.05 | 41.8 | |
| ReACTBackbone=Sonnet 4.52026.05 | 40.6 | |
| PIVOTBackbone=Haiku 4.5, Feedback regime=w/o HITL2026.05 | 40.6 | |
| SE-AgentBackbone=Sonnet 4.52026.05 | 40 | |
| Self-CritiqueBackbone=Q3 235B2026.05 | 37.6 | |
| SE-AgentBackbone=Q3 235B2026.05 | 37.5 | |
| ReACTBackbone=Haiku 4.52026.05 | 37 | |
| SE-AgentBackbone=Haiku 4.52026.05 | 37 | |
| PIVOTBackbone=Q3 235B, Feedback regime=w/o HITL2026.05 | 35.2 | |
| AgentDebugBackbone=Q3 235B2026.05 | 33.4 | |
| ReACTBackbone=Q3 235B2026.05 | 31.5 |