Home
Tongyi Lab Open-Sources LOGOS, First Unified Scientific Large Model Outperforming NatureLM with 1B Parameters
On June 18, 2026, Tongyi Lab announced the open-source release of LOGOS (Language Of Generative Objects in Science), the first multi-domain scientific generative foundation model built on a unified "scientific grammar."
Developed jointly by ATH-Token Foundry and the Huoqian Institute of Artificial Intelligence at Renmin University, this model aims to break the fragmented "one task, one expert model" status quo in traditional AI for Science (AI4S). By encoding diverse scientific objects—proteins, small molecules, materials, and chemical reactions—into a unified discrete token sequence, it enables cross-domain knowledge integration and autoregressive generation within a native large model framework.

LOGOS's core breakthrough lies in its innovative "scientific grammar" design and spatial interaction discretization technology. This allows the model to deeply understand complex 3D spatial interaction rules and perform autoregressive generation without relying on scarce 3D coordinate data or specialized geometric networks, ensuring full consistency in form and objective between pre-training and downstream tasks.
Evaluation results show that LOGOS-1B, with just 1 billion parameters, consistently matches or surpasses domain-specific methods across six representative tasks: pocket-conditioned ligand generation, retrosynthesis prediction (Top-1 accuracy 74.8%), pocket site identification (Top-n accuracy 58.5% on the HOLO4K dataset), and MOF material generation (a 76% increase in the proportion of new building units). In some tasks, it even outperformed NatureLM—which uses 8×7B parameters—while using only 1/56 of the parameter count.

Related article
OpenAI Unveils Codex Cloud Environments for Seamless Cross-Device Use
OpenAI has expanded its software engineering agent, Codex, with new capabilities that extend its utility beyond individual laptops. These updates introduce reusable cloud development environments, enabling access from any device.Announced during Open
Tencent WorkBuddy Launches on HarmonyOS PC App Market
On July 27, Zhang Jun, Tencent’s public relations director, announced the official launch of the company’s AI office assistant, WorkBuddy, on the HarmonyOS PC app market. This marks the debut of the first desktop AI agent within the HarmonyOS ecosyst
Alibaba DAMO Academy Unveils DAMO EAGLE AI for Non-Invasive Esophageal Cancer Detection
Alibaba has unveiled its latest cancer screening AI tool, DAMO EAGLE. Developed jointly by Alibaba DAMO Academy, Sichuan Cancer Hospital, and Sun Yat-sen University Cancer Center, this model detects esophageal cancer—including early-stage and precanc
Related Special Topic Recommendations
Comments (0)
0/500
On June 18, 2026, Tongyi Lab announced the open-source release of LOGOS (Language Of Generative Objects in Science), the first multi-domain scientific generative foundation model built on a unified "scientific grammar."
Developed jointly by ATH-Token Foundry and the Huoqian Institute of Artificial Intelligence at Renmin University, this model aims to break the fragmented "one task, one expert model" status quo in traditional AI for Science (AI4S). By encoding diverse scientific objects—proteins, small molecules, materials, and chemical reactions—into a unified discrete token sequence, it enables cross-domain knowledge integration and autoregressive generation within a native large model framework.

LOGOS's core breakthrough lies in its innovative "scientific grammar" design and spatial interaction discretization technology. This allows the model to deeply understand complex 3D spatial interaction rules and perform autoregressive generation without relying on scarce 3D coordinate data or specialized geometric networks, ensuring full consistency in form and objective between pre-training and downstream tasks.
Evaluation results show that LOGOS-1B, with just 1 billion parameters, consistently matches or surpasses domain-specific methods across six representative tasks: pocket-conditioned ligand generation, retrosynthesis prediction (Top-1 accuracy 74.8%), pocket site identification (Top-n accuracy 58.5% on the HOLO4K dataset), and MOF material generation (a 76% increase in the proportion of new building units). In some tasks, it even outperformed NatureLM—which uses 8×7B parameters—while using only 1/56 of the parameter count.

OpenAI Unveils Codex Cloud Environments for Seamless Cross-Device Use
OpenAI has expanded its software engineering agent, Codex, with new capabilities that extend its utility beyond individual laptops. These updates introduce reusable cloud development environments, enabling access from any device.Announced during Open
Tencent WorkBuddy Launches on HarmonyOS PC App Market
On July 27, Zhang Jun, Tencent’s public relations director, announced the official launch of the company’s AI office assistant, WorkBuddy, on the HarmonyOS PC app market. This marks the debut of the first desktop AI agent within the HarmonyOS ecosyst
Alibaba DAMO Academy Unveils DAMO EAGLE AI for Non-Invasive Esophageal Cancer Detection
Alibaba has unveiled its latest cancer screening AI tool, DAMO EAGLE. Developed jointly by Alibaba DAMO Academy, Sichuan Cancer Hospital, and Sun Yat-sen University Cancer Center, this model detects esophageal cancer—including early-stage and precanc











