InternScience releases Sage-A1.5 35B agentic model for scientific tasks
InternScience has published Sage-A1.5, a 35 billion‑parameter Mixture‑of‑Experts model built on Shanghai AI Laboratory’s Qwen3.6‑35B‑A3B. The model is packaged for Hugging Face Transformers and can be served with vLLM or SGLang, supporting a 200 K‑token context window, tool calling, and two optional modes – Agentic Orchestration and Verification.
Key points
- Sage-A1.5 is a 35B Mixture‑of‑Experts agentic model based on Qwen3.6‑35B‑A3B, released with Hugging Face weights.
- It scores 92.0 on XBench‑DS‑2510 and 95.1 on GAIA, outperforming larger frontier models on several benchmarks.
- The repo includes serving scripts for vLLM and SGLang, a 200K‑token context window, and optional orchestration and verification modes.
In the authors’ offline evaluation the model achieved top scores on several benchmarks, including 92.0 on XBench‑DS‑2510, 81.6 on IFBench, 90.0 on PinchBench, 81.0 on FrontierScience‑Olympiad, 48.3 on FrontierScience‑Research, 80.0 on CMT, 91.5 on OmniDocBench and a GAIA score of 95.1 that exceeds every larger model tested. Compared with its predecessor A1, Sage‑A1.5 improves SWE‑bench Pro by +17.6 points and several other agentic and engineering metrics.
The repository provides weights, configuration files, and example launch commands for both standard and speculative multi‑token prediction serving. Users can enable the orchestration mode by registering a spawnsubagent tool, and turn on verification via a chat‑template flag. The release is intended to give the community an open foundation for long‑horizon scientific problem solving.
The story so far
2 episodes →- InternScience releases Sage-A1.5 35B agentic model for scientific tasksthis story
InternScience/Sage-A1.5 [Finetune of Qwen 3.6 35b-a3b]
huggingface.co · 21 October 2026
Loading the full article…
This text was published by huggingface.co. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Nvidia expands open models for language, robotics, biology and quantum computing · 8 src
- OpenAI agents removed safety monitors and denied actions, researcher says · 12 src
- Google launches Gemini 4 Argon in Japan with up to 1 million-token context · 2 src
- Krafton releases Raon-1 mini preview, a 30B-A4B MoE model with 262,144 token context · 1 src
- OpenAI says its AI agents weakened internal misalignment monitoring system · 5 src
Comments
via GitHub Discussions