Anthropic Says Claude Opus 4.7 Beats NMR Software on 20 Compounds
Updated
Updated · Anthropic · Jun 3
Anthropic Says Claude Opus 4.7 Beats NMR Software on 20 Compounds
1 articles · Updated · Anthropic · Jun 3
Summary
20 post-cutoff ChemRxiv compounds formed the basis of Anthropic’s white paper, which found Claude Opus 4.7 matched or outperformed ChemDraw and MestReNova in 1D NMR prediction.
±0.079 ppm was Opus 4.7’s average hydrogen-error rate—best in the test—while its carbon predictions were effectively tied with MestReNova at ±1.37 versus ±1.48 ppm.
15 structure-elucidation problems showed the model could also work in reverse: it recovered all 8 simpler molecules on every attempt and solved 4 of 7 harder targets on all 3 runs.
Anthropic said the results suggest a general-purpose model can handle routine spectral analysis without chemistry-specific fine-tuning, though it still wants broader testing across hundreds of compounds, more scaffold classes and 2D NMR data.
The company is expanding its AI for Science chemistry work, aiming to help chemists with translation and analysis tasks that remain slow and hard to scale.
Does Claude’s success signal a true grasp of chemical principles, or is it merely a sophisticated pattern-matching feat?
How can a general AI guarantee precision in chemistry when 'hallucinations' pose a foundational risk to drug discovery?
Can a generalist AI truly threaten specialized drug discovery firms whose key advantage is proprietary experimental data?
Claude Opus 4.7: Transforming NMR Analysis with Record-Low 0.079 ppm Error and Advanced Visual AI
Overview
Claude Opus 4.7 represents a major leap in general-purpose AI, especially for scientific fields like NMR analysis. Its improved multimodal support allows it to process and interpret complex visual information, such as intricate diagrams and spectral data, more effectively than before. With superior vision capabilities, Claude Opus 4.7 can handle high-resolution images up to 2,576 pixels, making it easier for scientists to extract detailed data from various formats. These advancements streamline scientific workflows, enabling faster and more accurate analysis, and mark a significant step forward in how AI can support specialized research tasks.