Shanghai Artificial Intelligence Laboratory's Sage S2.1 Flash Tops MLE-bench Lite
·2026.11.16 20:36
Key point
The model supports over ten scientific input types and ranks first on four of fifteen benchmarks.
Details
- Sage S2.1 Flash is a scientific multimodal model from Shanghai Artificial Intelligence Laboratory designed for research tasks requiring multiple rounds of analysis and verification.
- It achieved a score of 90.3 on MLE-bench Lite, placing first ahead of DeepSeek-V4.1-Flash-1017 at 89.2.
- The model placed first on OfficeQA Pro (72.2), Harvey LAB (25.0), and SpreadsheetBench 2 (37.1).
- It placed second on seven benchmarks including AutomationBench, ALE-CLI, BFCL V4, and OSWorld-Verified.
Architecture and Capabilities
- Uses the unified representation architecture of Sage SciPerception to support more than ten types of scientific input.
- Encodes features from molecular graphs, crystal structures, and protein sequences into a shared representation space for reasoning alongside natural language instructions.
- Supports tasks such as prediction, classification, site localization, quantitative analysis, and structure inference.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.