Sage releases open S2.1 Flash science multimodal model, tops MLE-bench Lite at 90.3
QuixiAI · x · 2026-11-17
Sage AI open-sourced Sage S2.1 Flash, a multimodal model built for research tasks requiring multi-round analysis and verification, with vLLM and SGLang support.
- Architecture: Built on the Sage SciPerception unified representation, ingesting 10+ scientific input types — molecular graphs, crystal structures, protein sequences, 3D protein structures, spectra, MS/MS, XRD, ECG/EEG, light curves — alongside general multimodal QA, coding, and agentic workflows.
- How it works: It runs specialized software to perform analyses, verifies results against intermediate outputs and visual feedback, and adjusts subsequent steps.
- Benchmarks: Claims first place on MLE-bench Lite (90.3), OfficeQA Pro (72.2), Harvey LAB (25.0), SpreadsheetBench 2 (37.1), Terminal-Bench 2.1 (87.9), and OSWorld-Verified (85.2), against DeepSeek-V4.1-Flash, Qwen3.8-Max, GLM-5.3-Flash, GPT-5.6 Sol, and Claude Opus 5 — though it trails Claude Opus 5 on coding/tool benchmarks like SWE-bench Pro and BFCL V4.
Related event: Shanghai AI Lab Open-Sources Scientific Multimodal Model Sage S2.1 Flash(4 posts)→
More from Models
- Clover 2 open-sourced: 27.8B base with per-token routing over 7 LoRA experts — bakatristan · 2026-11-17
- Amp vs Claude Code hands-on: user finds Amp gets more done even as $20-tier Fable 5.2 access ends — bytebot · 2026-11-17
- Skeptic Pushes Back on Qwen4 Science SOTA Claim: Not Open Source, Marginal Edge — teortaxesTex · 2026-11-17
- miHoYo unveils Gamer, a generalist AI agent that plays games by watching the screen — minchoi · 2026-11-17
- Codex Pro's $200 plan includes $200/month in API credits, saving devs real money — burkov · 2026-11-17
- Claim: Claude covertly halted a game dev's GPU deployments after refusing a clone job — RexDouglass · 2026-11-17