AVG
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Question Answering | AVG. | EM47.5 | 48 | |
| Speculative Decoding | Avg. | Speedup6.73 | 32 | |
| Mathematical and Scientific Reasoning | AVG GSM8K MATH-500 AIME 2024 OlympiadBench GPQA Diamond | Accuracy78.4 | 24 | |
| Mathematical Reasoning | Avg. (held out) | Accuracy62.8 | 24 | |
| change detection | Avg across SYSU, LEVIR, GVLM, CLCD, OSCD | Precision84.8 | 23 | |
| General Reasoning | AVG Reasoning Suite | Accuracy77.4 | 18 | |
| Low-Light Image Enhancement | AVG. DICM, MEF, LIME, NPE, VV | NIQE3.589 | 17 | |
| Document Reranking | AVG | NDCG@554.373 | 14 | |
| Reasoning Performance (Aggregate) | AVG | TPF351 | 14 | |
| Question Answering | AVG. Aggregate of NQ, TQA, HQA, 2WIKI (test) | EM42.5 | 14 | |
| Reasoning | AVG AMC 2023, AIME 2025, HumanEval, LiveCodeBench | Accuracy83.4 | 12 | |
| Emotion Recognition in Conversation | AVG IEMOCAP, MELD, EmoryNLP | W-F160.91 | 11 | |
| Selective classification | Avg (all) | AURC (10^-2 Scale)0.215 | 11 | |
| Selective classification | Avg 1K | AURC (Scale 10^-2)0.248 | 11 | |
| Geolocation | AVG (test) | City Acc (25km)8.3 | 10 | |
| Detection | AVG | AUC0.897 | 10 | |
| Sparse-view reconstruction | Avg. across four cases Combined (test) | PSNR40.63 | 9 | |
| Multi-task Language Understanding | AVG Across All Benchmarks | Throughput12.89 | 8 | |
| Scene Text Recognition | AVG 12 benchmarks | Word Accuracy91.33 | 8 | |
| Image Classification | Avg-14 | Top-1 Clean Accuracy76.19 | 7 | |
| Category-level pose estimation | Avg. All five datasets | Median Error (Med)25.5 | 5 | |
| Preference Alignment | Avg. | GRA70.6 | 4 | |
| Machine Translation | Avg (News, Flores, Subtitle, Travel) German-English (test aggregate) | DA87.77 | 4 |
Showing 23 of 23 rows