Sage-A1.5: Shanghai AI Lab's 35B agentic MoE that beats frontier models on scientific research
Shanghai AI Laboratory (InternScience) released Sage-A1.5, a 35B-parameter mixture-of-experts model based on Qwen3.6, built for long-horizon agentic work and open scientific discovery. It adds agentic orchestration (parallel subagents with isolated contexts via a spawn_subagent tool) and an opt-in verification mode where the model pauses mid-trajectory to review and revise its own plan without an external critic.
Why it matters
A mid-size open-weight model claiming to match or beat frontier systems (Kimi-K3, GLM-5.3, GPT-5.6-sol) on scientific research — GAIA 95.1, XBench-DS-2510 92.0 — with agentic orchestration designed into the model card rather than bolted on as a harness.
What you could build with it
A research team or startup could wire Sage-A1.5's Agentic Orchestration mode (spawn_subagent tool) into an open-science assistant: a literature-review agent that fans out independent sub-investigations across papers, datasets, and codebases in parallel, then synthesizes a verifiable report with inspectable artifacts. The model's opt-in verification mode lets it pause mid-trajectory to review and revise its plan without an external critic model.
Does it hold up?
Too early to judge: the weights landed on Hugging Face about a day ago and the only evidence so far is the lab's own benchmark table — GAIA 95.1, BrowseComp 82.8 with orchestration — with no independent reproductions yet.
Built with Sage-A1.5
- InternScience/Sage-A1.5Hugging Face · Official model card: Sage-A1.5, the 35B agentic MoE based on Qwen3.6 for scientific research (245 likes).
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.