National University of Singapore
We track 18 benchmarks that National University of Singapore made or helped make.
1 in use 17 with public code 8 with a leaderboard here
Benchmarks
| Name | Kind | Released | Best score | Status |
|---|---|---|---|---|
| MixEval Models answer real user queries matched to similar questions from existing benchmarks. Ground-truth answers make it a cheap proxy for Chatbot Arena rankings. | Overall chat quality | Jun 2024 | 89.9% | Dropped |
| NExT-QA Models answer causal, temporal and descriptive questions about short videos of daily life. Labs usually report the multiple choice split. | Video understanding | May 2021 | | Dropped |
| Last Secure Code Benchmark Agents rebuild code that a real vulnerability once affected, from a specification that never mentions security. Hidden tests check function and replay an exploit. | Secure code and review | Oct 2026 | 25.1% | In use |
| TAT-DQA Models answer questions about financial report pages that mix tables and text, often needing arithmetic. Scored by exact match and F1. | Documents, charts and text in images | Jul 2022 | | Dropped |
| General-Bench Models perform multimodal understanding and generation tasks. General-Level averages task scores and measures performance above specialist baselines at progressively broader levels. | General visual questions | May 2025 | | Not in cards |
| ViMU Models explain what a short video implies and select the cues that support their reading. They also identify rhetorical devices and social signals. | Video understanding | May 2026 | 46.9% | Not in cards |
| Learn2Play Bench Agents play unfamiliar games whose rules are not given and must learn from experience across episodes. Improvement over repeated play is the measure. | Puzzles and chess | Oct 2026 | 81.8 | Not in cards |
| EgoBlind Models answer questions from blind users using only video recorded before each question. GPT-4o mini checks each answer against human references. | Video understanding | Mar 2025 | 59.3% | Not in cards |
| CutVerse Agents complete editing tasks in seven applications, including Premiere Pro, Photoshop, and DaVinci Resolve. Vision models grade progress against milestones in the recorded workflow. | Desktop apps | May 2026 | 68.3% | Not in cards |
| RSIBench-Data Coding agents act as data-centric researchers. With a fixed Tinker post-training stack and budget, they refine training data from dev-benchmark feedback, and the final model is scored on held-out test benchmarks. | Machine learning engineering | Jul 2026 | 32.6% | Not in cards |
| SO-Arena Models choose answers across 11 domains and five capability groups. Evaluation tests accuracy, stability under changed option sets, and recognition that the correct answer is missing. | Language reasoning | Nov 2026 | 82.3% | Not in cards |
| WorldGUI Desktop GUI agents do 611 tasks that start from varied initial states, such as the target app not being open. Tasks come with instructional videos and project files. | Desktop apps | Feb 2025 | 45.8% | Not in cards |
| EndoVis18-VQA Models answer questions about organs, tools, tool positions, and interactions in images from robotic kidney surgery. The benchmark accepts class labels or sentence answers in separate tasks. | Medicine | Jun 2022 | Not in cards | |
| MultiRef-Compass Models generate audio-video content from several visual references and text instructions across 350 samples. Fourteen metrics assess quality, reference fidelity, audio-visual consistency, and instruction following. The evaluator combines automatic measurements with multimodal model judgments. | Video generation | Jul 2026 | Not in cards | |
| MyCareersFuture Skill Retrieval Models rank required skills from job descriptions on Singapore’s government employment portal. Employer tags supply the answers, including skills the description leaves implicit. | Other professions | Dec 2020 | Not in cards | |
| PiRhDy Accompaniment Suggestion Models rank 50 accompaniment phrases for a MIDI melody. Several tracks can belong with the melody. The authors report mean average precision. | Audio understanding and voice | Oct 2020 | Not in cards | |
| PiRhDy Melody Completion Models rank 50 possible continuations of a MIDI melody. One phrase follows the prompt in the original piece. The authors report mean average precision. | Audio understanding and voice | Oct 2020 | Not in cards | |
| TAT-QA Models answer 16,552 questions about 2,757 passages from real financial reports that mix a table with text. Many answers need arithmetic, counting or comparison. Scored by exact match and F1. | Finance | May 2021 | Not in cards |