Loading the SOTA2 catalog…
SOTA2 Research · benchmarks
Compare state-of-the-art methods across the tasks and datasets used to measure AI progress.
| Task | Dataset | Domain | Trend | Results | Last Update |
|---|---|---|---|---|---|
| Expert-level Question Answering | GPQA Diamond | 39 | Jun 1, 2026 | ||
| Reading Comprehension | Belebele | 39 | Mar 10, 2026 | ||
| Code Generation | HumanEval OOD | 39 | May 15, 2026 | ||
| Visual document retrieval | ViDoRe V2 | 39 | Jun 1, 2026 | ||
| Automatic Speech Recognition | LS Clean |
| 39 |
| Jun 9, 2026 |
| Text-to-speech | Seed-TTS (eval) | 39 | Feb 26, 2026 |
|---|
| Mathematical Reasoning | HMMT Nov 2025 | 39 | Jul 9, 2026 |
|---|
| Long-context language understanding | RULER 32k context length | 39 | May 13, 2026 |
|---|
| Machine Translation | FLORES-200 (devtest) | 39 | May 28, 2026 |
|---|
| Transferable Adversarial Attack | AdvBench LLM Classifier (test) | 39 | Feb 26, 2026 |
|---|
| Machine-Generated Text Detection | Essay (test) | 39 | May 18, 2026 |
|---|
| Open ended generation | TruthfulQA Without Rejected Samples open-ended (full) | 39 | Feb 26, 2026 |
|---|
| Open ended generation | TruthfulQA With All Samples open-ended (full) | 39 | Feb 26, 2026 |
|---|
| Automated Software Engineering | SWE-bench Verified | 39 | Feb 26, 2026 |
|---|
| Mathematical Reasoning | GSM8K (test) | 39 | Feb 26, 2026 |
|---|
| Visual Retrieval-Augmented Generation | Visual-RAG | 39 | May 14, 2026 |
|---|
| Mathematical Reasoning | AIME 24 | 39 | Feb 26, 2026 |
|---|
| NER | Malayalam NER (test) | 39 | Feb 26, 2026 |
|---|
| Sentiment Classification | Bangla (test) | 39 | Feb 26, 2026 |
|---|
| Mathematical Reasoning | OlympiadBench | 39 | Feb 26, 2026 |
|---|
| Conditional Shapley value estimation | Abalone M=8 (all) | 39 | Feb 26, 2026 |
|---|
| Safety evaluation | UnsafeBench | 39 | Jul 2, 2026 |
|---|
| Medical Visual Question Answering | MMMU Health & Medicine (test) | 39 | Apr 14, 2026 |
|---|
| Text Clustering | DBp F | 39 | Feb 26, 2026 |
|---|
| Conformal Prediction | 15 datasets (average) | 39 | May 25, 2026 |
|---|