FlashNEWS BRIEFING

Shanghai AI Lab Open-Sources Scientific Multimodal Large Model InternLM-S2.1 Flash

1 min read

88Z AI, 17 November — Shanghai AI Laboratory has open-sourced the scientific multimodal large model InternLM-S2.1 Flash. Built on the Unified Representation Architecture, it supports inputs such as molecules, spectra, and biological sequences. It combines scientific understanding with agent capabilities to directly control research software, reducing inference costs to one-tenth of InternLM-Xinghe for the same tasks. Featuring a recursive self-improvement mechanism, it achieved first place in four benchmarks and seco

At a glance

  • Supports diverse scientific data inputs and direct control of research software, with inference costs reduced to one-tenth of InternLM-Xinghe
  • Features a recursive self-improvement mechanism, achieving first place in four and second in seven out of fifteen general agent benchmarks
Cover image for Shanghai AI Lab Open-Sources Scientific Multimodal Large Model InternLM-S2.1 Flash
Cover image for Shanghai AI Lab Open-Sources Scientific Multimodal Large Model InternLM-S2.1 Flash

The update

The Shanghai Artificial Intelligence Laboratory has officially open-sourced InternLM-S2.1 Flash, a scientific multimodal large model. Built on the Unified Representation Architecture, it processes diverse scientific inputs including molecular structures, spectral data, and biological sequences. The model combines deep scientific understanding with agent technologies, enabling direct operation of various research software to assist in scientific work. In terms of efficiency, it significantly reduces inference costs to just one-tenth of those incurred by InternLM-Xinghe for identical tasks. Furthermore, InternLM-S2.1 Flash incorporates a recursive self-improvement mechanism, granting it self-evolution capabilities. In performance evaluations, the model demonstrated strong comprehensive intelligence, securing first place in four benchmarks and second place in seven out of fifteen general agent benchmark tests.

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.

  1. 书生SageSource published

Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.

Related reading

All stories
88Z.ai