Performance Estimation on GQA (MAE)
0MAEBQ-RPF
Evaluation Results
| Method | Links | |
|---|---|---|
| BQ-RPFBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Active Selection + BQ2026.04 | 0 | |
| BQ-SF RoundedBudget=1%, Target Model=GPT-4o, Selection Strategy=Active Selection + BQ2026.04 | 0.002 | |
| BQ-SF RoundedBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Active Selection, Integer Rounding=true2026.04 | 0.002 | |
| BQ-SF RoundedBudget=1%, Target Model=Gemini 3 Flash, Selection=Active Selection + BQ2026.04 | 0.002 | |
| BQ-TPFBudget=1%, Model=Gemma 3 27B, Selection Strategy=Active Selection2026.04 | 0.003 | |
| BQ-SF RoundedTarget Model=Claude 4.5 Opus, Budget=1%, Selection Strategy=Active Selection + BQ2026.04 | 0.004 | |
| BQ-RPF RoundedBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Active Selection2026.04 | 0.005 | |
| BQ-TPFBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Active Selection + BQ2026.04 | 0.007 | |
| BQ-RPFSelection Mode=Active Selection, Budget=1%, Target Model=GPT-52026.04 | 0.007 | |
| BQ-SF RoundedTarget Model=GPT-5.2, Budget=1%, Selection Strategy=Active Selection2026.04 | 0.008 | |
| BQ-RPF RoundedBudget=1%, Model=Gemma 3 27B, Selection Strategy=Active Selection2026.04 | 0.011 | |
| BQ-SFBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Active Selection + BQ, Rounding=false2026.04 | 0.013 | |
| BQ-SFSelection Strategy=Active Selection, Feature Type=Score Features (SF), Rounding=True, Labeling Budget=1%2026.04 | 0.013 | |
| BQ-SFScenario=New Model (NM), Feature Type=Score Features (SF), Rounding=True, Labeling Budget=1%2026.04 | 0.013 | |
| BQ-TPFTarget Model=GPT-5.2, Budget=1%, Selection Strategy=Active Selection2026.04 | 0.016 | |
| BQ-RPF RoundedSelection Mode=Active Selection, Budget=1%, Target Model=GPT-52026.04 | 0.016 | |
| BQ-SFBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Active Selection2026.04 | 0.019 | |
| BQ-SFModel=Gemma 3 12B, Budget=1%, Selection Strategy=Active Selection2026.04 | 0.02 | |
| BQ-RPF RoundedBudget=1%, Target Model=GPT-4o, Selection Strategy=Active Selection + BQ2026.04 | 0.021 | |
| BQ-RPF RoundedBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Active Selection, Integer Rounding=true2026.04 | 0.021 | |
| Random SamplingTarget Model=Claude 4.5 Opus, Budget=1%2026.04 | 0.022 | |
| BQ-SF RandBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Random2026.04 | 0.022 | |
| BQ-SF RandBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Random Selection + BQ2026.04 | 0.023 | |
| BQ-SFTarget Model=Claude 4.5 Opus, Budget=1%, Selection Strategy=Active Selection + BQ2026.04 | 0.023 | |
| BQ-SF RandModel=Gemma 3 12B, Budget=1%, Selection Strategy=Random Selection2026.04 | 0.023 | |
| BQ-SF RandBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Random Selection2026.04 | 0.024 | |
| BQ-SFBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Active Selection2026.04 | 0.024 | |
| BQ-RPF RoundedBudget=1%, Target Model=Gemini 3 Flash, Selection=Active Selection + BQ2026.04 | 0.024 | |
| BQ-TPFBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Active Selection + BQ, Rounding=false2026.04 | 0.025 | |
| BQ-SFBudget=1%, Target Model=Gemini 3 Flash, Selection=Active Selection + BQ2026.04 | 0.025 | |
| BQ-SFBudget=1%, Target Model=GPT-4o, Selection Strategy=Active Selection + BQ2026.04 | 0.026 | |
| BQ-RPF RandBudget=1%, Target Model=Gemini 3 Flash, Selection=Random Selection + BQ2026.04 | 0.026 | |
| BQ-RPFBudget=1%, Target Model=GPT-4o, Selection Strategy=Active Selection + BQ2026.04 | 0.027 | |
| BQ-SFBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Active Selection + BQ2026.04 | 0.03 | |
| BQ-RPF RoundedBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Active Selection + BQ, Rounding=true2026.04 | 0.031 | |
| RF+LUREBudget=1%, Target Model=Claude 4.5 Sonnet2026.04 | 0.031 | |
| LR+LUREModel=Claude 3.7 Sonnet, Budget=1%2026.04 | 0.032 | |
| BQ-SF RoundedSelection Mode=Active Selection, Budget=1%, Target Model=GPT-52026.04 | 0.033 | |
| LR+ISBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Random2026.04 | 0.034 | |
| BQ-SFTarget Model=GPT-5.2, Budget=1%, Selection Strategy=Active Selection2026.04 | 0.035 | |
| BQ-RPFModel=Claude 3.7 Sonnet, Budget=1%, Selection=Active2026.04 | 0.036 | |
| BQ-SF RoundedBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Active Selection + BQ, Rounding=true2026.04 | 0.038 | |
| BQ-RPF RandTarget Model=Claude 4.5 Opus, Budget=1%, Selection Strategy=Random Selection + BQ2026.04 | 0.038 | |
| BQ-SF RandBudget=1%, Target Model=Gemini 3 Flash, Selection=Random Selection + BQ2026.04 | 0.038 | |
| BQ-TPF RandBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Random Selection + BQ2026.04 | 0.039 | |
| BQ-SF RoundedBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Active Selection + BQ2026.04 | 0.039 | |
| BQ-TPF RandBudget=1%, Target Model=Gemini 3 Flash, Selection=Random Selection + BQ2026.04 | 0.039 | |
| BQ-SF RoundedModel=Gemma 3 12B, Budget=1%, Selection Strategy=Active Selection2026.04 | 0.039 | |
| BQ-RPFBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Active Selection2026.04 | 0.04 | |
| LR+LUREBudget=1%, Target Model=GPT-52026.04 | 0.041 | |
| BQ-SF RandBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Random Selection + BQ2026.04 | 0.042 | |
| BQ-RPFBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Active Selection2026.04 | 0.042 | |
| BQ-RPF RandBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Random Selection + BQ2026.04 | 0.044 | |
| RF+LUREBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Standard2026.04 | 0.044 | |
| BQ-RPF RoundedBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Active Selection + BQ2026.04 | 0.044 | |
| BQ-RPF RandSelection Mode=Random Selection, Budget=1%, Target Model=GPT-52026.04 | 0.044 | |
| BQ-RPF RandTarget Model=GPT-5.2, Budget=1%, Selection Strategy=Random Selection2026.04 | 0.045 | |
| BQ-SF RoundedBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Active Selection2026.04 | 0.045 | |
| BQ-RPF RandBudget=1%, Target Model=GPT-4o, Selection Strategy=Random Selection + BQ2026.04 | 0.046 | |
| Random SamplingBudget=1%, Model=Gemini 2.5 Pro2026.04 | 0.047 | |
| Random SamplingBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Random Selection2026.04 | 0.05 | |
| BQ-TPF RandTarget Model=GPT-5.2, Budget=1%, Selection Strategy=Random Selection2026.04 | 0.05 | |
| BQ-SFBudget=1%, Model=Gemma 3 27B, Selection Strategy=Active Selection2026.04 | 0.051 | |
| BQ-TPF RandModel=Claude 3.7 Sonnet, Budget=1%, Selection=Random2026.04 | 0.051 | |
| BQ-SFSelection Mode=Active Selection, Budget=1%, Target Model=GPT-52026.04 | 0.051 | |
| BQ-RPFBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Active Selection + BQ, Rounding=false2026.04 | 0.052 | |
| BQ-SF RandTarget Model=GPT-5.2, Budget=1%, Selection Strategy=Random Selection2026.04 | 0.053 | |
| LR+ISBudget=1%, Target Model=Claude 4.5 Sonnet2026.04 | 0.053 | |
| BQ-RPF RandBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Random Selection + BQ2026.04 | 0.054 | |
| BQ-TPFBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Active Selection2026.04 | 0.054 | |
| RF+ISBudget=1%, Target Model=Gemini 3 Flash2026.04 | 0.054 | |
| BQ-SF RandBudget=1%, Target Model=GPT-4o, Selection Strategy=Random Selection + BQ2026.04 | 0.055 | |
| Random SamplingBudget=1%, Target Model=Claude 4.5 Sonnet2026.04 | 0.055 | |
| BQ-SF RandTarget Model=Claude 4.5 Opus, Budget=1%, Selection Strategy=Random Selection + BQ2026.04 | 0.055 | |
| BQ-SFModel=Claude 3.7 Sonnet, Budget=1%, Selection=Active2026.04 | 0.055 | |
| BQ-TPF RandBudget=1%, Model=Gemini 2.5 Pro2026.04 | 0.055 | |
| Random SamplingBudget=1%, Model=Gemma 3 27B2026.04 | 0.056 | |
| LR+LUREBudget=1%, Target Model=Claude 4.5 Sonnet2026.04 | 0.056 | |
| BQ-TPF RandBudget=1%, Target Model=Claude 4.5 Sonnet, Selection Strategy=Random Selection2026.04 | 0.056 | |
| RF+LUREBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Random2026.04 | 0.056 | |
| BQ-SF RandSelection Mode=Random Selection, Budget=1%, Target Model=GPT-52026.04 | 0.057 | |
| LR+LUREBudget=1%, Target Model=Gemini 3 Flash2026.04 | 0.057 | |
| BQ-RPFBudget=1%, Target Model=Gemini 3 Flash, Selection=Active Selection + BQ2026.04 | 0.058 | |
| BQ-RPF RandBudget=1%, Model=Gemma 3 27B, Selection Strategy=Random Selection2026.04 | 0.059 | |
| BQ-SF RoundedBudget=1%, Model=Gemma 3 27B, Selection Strategy=Active Selection2026.04 | 0.059 | |
| BQ-TPF RandBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Random Selection + BQ2026.04 | 0.059 | |
| RF+LURETarget Model=Claude 4.5 Opus, Budget=1%2026.04 | 0.06 | |
| LR+LUREBudget=1%, Target Model=Claude 3.5 Haiku, Selection Strategy=Standard2026.04 | 0.062 | |
| BQ-RPF RandBudget=1%, Target Model=Gemini 3 Pro, Selection Strategy=Random2026.04 | 0.062 | |
| BQ-RPF RoundedTarget Model=Claude 4.5 Opus, Budget=1%, Selection Strategy=Active Selection + BQ2026.04 | 0.064 | |
| LR+ISBudget=1%, Target Model=Gemini 3 Flash2026.04 | 0.064 | |
| RF+ISBudget=1%, Model=Gemma 3 27B2026.04 | 0.065 | |
| BQ-TPF RandBudget=1%, Target Model=GPT-4o, Selection Strategy=Random Selection + BQ2026.04 | 0.066 | |
| LR+ISBudget=1%, Target Model=GPT-4o, Selection Strategy=Random Selection2026.04 | 0.067 | |
| LR+LUREBudget=1%, Model=Gemma 3 27B2026.04 | 0.069 | |
| LR+ISTarget Model=Claude 4.5 Opus, Budget=1%2026.04 | 0.069 | |
| BQ-SF RoundedModel=Claude 3.7 Sonnet, Budget=1%, Selection=Active2026.04 | 0.069 | |
| BQ-RPF RandModel=Gemma 3 12B, Budget=1%, Selection Strategy=Random Selection2026.04 | 0.069 | |
| LR+ISBudget=1%, Evaluated Model=GPT-5.1, Selection Strategy=Random Selection2026.04 | 0.07 | |
| RF+LUREModel=Gemma 3 12B, Budget=1%2026.04 | 0.07 |