88Z AI, 15 September — Shanghai AI Laboratory has released ShuSheng-S2, an open-source multimodal foundation model. Its general capabilities rank among the top tier of open-source models. The model performs comparably to leading closed-source systems in scientific long-horizon tasks such as biology, materials science, chemistry, and advanced mathematics. By employing a Memory Decoder architecture with pluggable professional memory modules, ShuSheng-S2 balances domain expansion with general proficiency while strengthen
At a glance
- ShuSheng-S2 ranks top-tier among open-source models, matching leading closed-source systems in scientific long-horizon tasks
- Uses Memory Decoder architecture with pluggable memory modules to enhance long-horizon reasoning and agent execution
The update
Shanghai AI Laboratory has officially open-sourced ShuSheng-S2, a multimodal foundation model. The model demonstrates excellent general capabilities, securing a position in the first tier of open-source systems. It matches the performance of leading closed-source models in scientific long-horizon tasks, including biology, materials science, chemistry, IMO-Proof, and AdvancedMathBench. Technically, ShuSheng-S2 introduces a Memory Decoder architecture featuring pluggable professional memory modules. This design effectively balances domain-specific expansion with general foundational capabilities. It also enhances stability in long-horizon reasoning chains and improves execution efficiency for agent tasks, providing robust support for complex scientific computing and automation.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 书生InternSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.