Sage-A1.5: From Long-Horizon Execution to Scientific Exploration with Capability Composition
Wenjie Lou, Tianshuo Peng, Shiyang Feng, Shengji Tang, Runmin Ma, Ruizhe Chen, Jinxin Shi, Haoyang Peng and 91 others
Shanghai Artificial Intelligence Laboratory
Published October 20, 2026 · on the feed October 21, 2026 · ▲ 72 · arXiv 2610.28578
The abstract
In this report, we introduce Sage-A1.5, a 35B Mixture-of-Experts agentic model that builds on A1 to advance long-horizon problem solving through adaptive capability composition. Compared with A1, Sage-A1.5 learns not only to solve individual tasks, but also to organize work and revise its approach during execution. Sage-A1.5 highlights three improvements: agentic orchestration, verification, and open scientific discovery. First, agentic orchestration extends single-agent execution through task delegation and parallel collaboration among subagents with isolated contexts, enabling complex tasks to be decomposed and addressed concurrently and supporting efficient execution while reducing interference across concurrent subagents. Second, agentic verification strengthens long-horizon reliability through progress checking, evidence validation, and reflective strategy revision, allowing the agent to recognize unproductive search paths and pursue alternative approaches. Third, open scientific discovery extends agentic capabilities beyond answering predefined questions to conducting exploratory research, connecting scientific inquiry with interactive analysis and inspectable research artifacts. We showcase these capabilities through complicated agentic tasks and multidisciplinary scientific applications. Besides, we hope Sage-A1.5 provides the community with an accessible foundation for open scientific discovery, offering a practical path from long-horizon task execution to collaborative, verifiable, and inspectable scientific exploration.
The Brief
2:34A short podcast summary of the paper. Written and voiced by AI.
Transcript
Sage-A1.5. From Long-Horizon Execution to Scientific Exploration with Capability Composition. By Wenjie Lou and ninety-eight others, from the Shanghai Artificial Intelligence Laboratory. Published on the twentieth of October. Its code repository is worked out: about five hundred and fifty source files, with a readme and a licence.
Solving a long-horizon problem takes more than solving each step in turn. The work has to be organised, and when an approach is not paying off, it has to be revised while it is still under way. This report sets out to move an agentic model beyond solving individual tasks toward that kind of self-directed work.
Sage-A1.5 is a Mixture-of-Experts agentic model of thirty-five billion parameters, built on its predecessor, A1. The report describes its advance as adaptive capability composition. Compared with A1, it learns not only to solve individual tasks but also to organise its work and revise its approach during execution. Three improvements carry this: agentic orchestration, agentic verification, and open scientific discovery.
Orchestration extends what a single agent does on its own. Tasks can be delegated, and subagents can collaborate in parallel, each with its own isolated context. A complex task can then be broken apart and its pieces addressed concurrently. That supports efficient execution, and keeping the contexts isolated reduces interference between subagents running at the same time.
Verification aims at reliability over a long horizon. It combines checking progress, validating evidence, and revising strategy reflectively. Together these let the agent recognise when a search path is not productive, and turn to an alternative approach instead of pressing on.
Open scientific discovery pushes the agent past answering predefined questions, toward conducting exploratory research. It connects scientific inquiry with interactive analysis, and with research artifacts that can be inspected.
The report showcases these capabilities through complicated agentic tasks and through scientific applications spanning several disciplines, the two settings the title names: long-horizon execution and scientific exploration.
Composition is the word doing the work in the title: delegating, checking and exploring are framed as capabilities brought together adaptively, rather than as a single new skill. The authors hope Sage-A1.5 gives the community an accessible foundation for open scientific discovery, and a practical path from executing long tasks to scientific exploration that is collaborative, verifiable and inspectable.
Paper Quest · 7 chapters · about 38 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
Solving each step is not the same as finishing the job. Here is where long work breaks.
Your agent model solves single tasks well, step by step. But real projects in software, machine learning and science mean many related subtasks, comparing approaches and changing course as evidence arrives.
The road ahead
- 1 · The problem ~7 min An agent that solves single tasks step by step stalls on real projects, where many related subtasks must be organised and conclusions checked.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min Sage-A1.5 learns, inside the model, to split work among parallel subagents and to check and revise its own progress, with domain skills merged from teachers.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min Start from A1, delegate subtasks to parallel subagents, check progress and evidence, and switch strategy when a search path stops paying off.
- 6 · What they found ~3 min
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.