Shanghai AI Laboratory releases 35B agentic model Sage-A1.5
Original title: InternScience/Sage-A1.5 — new model trending #30 on Hugging Face
Weights are on Hugging Face for Transformers, vLLM and SGLang; the lab reports SOTA on several agent and science benchmarks.
Shanghai AI Laboratory published Sage-A1.5, a 35B mixture-of-experts model built on A1 for long-horizon scientific problem solving. The model card describes agentic orchestration across subagents, an opt-in verification mode, tool use and open scientific discovery. The lab reports competitive scores against larger systems and large gains versus A1 on engineering and agentic workloads.
Unconfirmed: Benchmark rankings are the lab's own comparisons, not independently verified here.
- The card says Sage-A1.5 is about 35B parameters and a mixture of experts.
- Reported scores include 92.0 on XBench-DS-2510 and 95.1 on GAIA.
- Versus A1, SWE-bench Pro is reported up 17.6 and MLE-Lite up 21.3.
- Artifacts are compatible with Hugging Face Transformers, vLLM and SGLang.
- Organizations
- Shanghai AI Laboratory
- AI models
- Sage-A1.5
From the original article
9 min read at huggingface.cotext-generation · 58 downloads · 160 likes · license apache-2. 0
By InternScience · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.