Is Sage-A1.5 state of the art?Pending benchmarks.Announced Oct 21, 2026; no independent results yet.
An open agentic model for long-horizon scientific work from InternScience at Shanghai AI Laboratory, fine-tuned from Qwen3.6-35B-A3B. It adds opt-in parallel subagents (Agentic Orchestration) and an opt-in in-reasoning self-review (Verification Mode), with native tool use and a 262K context. Apache 2.0.
The details so far
- Lab
- InternScience (Shanghai AI Laboratory)
- Announced
- Oct 21, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- About 36B total, about 3B active (Qwen3.6-35B-A3B MoE)
- Base model
- Qwen3.6-35B-A3B
- Context
- 262,144 tokens
- Inputs
- Text and images (includes a vision encoder; a text-only serving mode skips it)
- Modes
- Agentic Orchestration (parallel subagents via a spawn_subagent tool) and Verification Mode (chat-template flag); both off by default
- Predecessor
- Builds on InternScience's Agents-A1
- Formats
- BF16, FP8 and GGUF (F16, Q8_0, Q4_K_M); Transformers, vLLM and SGLang
- License
- Apache 2.0
- Hugging Face
- Repo created Oct 15, 2026 (UTC); FP8 and GGUF builds and the ModelScope announcement on Oct 21
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: 90.0 on PinchBench, 81.0 on FrontierScience-Olympiad and 48.3 on FrontierScience-Research, ahead of Kimi-K3, GLM-5.3 and GPT-5.6-sol in InternScience's comparison.
Lab claim: overall best in its comparison on XBench-DS-2510 (92.0), IFBench (81.6), CMT (80.0) and OmniDocBench (91.5), and 95.1 on GAIA.
Sources
REPOSage-A1.5 on Hugging Facehuggingface.co/InternScience/Sage-A1.5 · Oct 21, 2026LABSage-A1.5 project pagealpha-innovator.github.io/Sage-A1.5/ · Oct 21, 2026PRESSModelScope announcement on Xx.com/ModelScope2022/status/2112816255375745132 · Oct 21, 2026Tracking continues until an independent board lists it.