Artificial Analysis launches the Intelligence Index v5, with new coding and science benchmarks, built on private datasets; Claude Opus 5.5 (max) leads at 58

AI model benchmarks are now built on private datasets, meaning performance metrics may become less comparable across vendors.

Artificial Analysis has released its Intelligence Index v5, a benchmark for AI model capabilities. This new version incorporates two new benchmarks and three upgrades, with a significant portion of its evaluation datasets now being private. In the Index, Anthropic's Claude Opus 5.5 (max) achieved the top score of 58. The release also highlights Cartesia Managed Agents, which scored 1115 on the Speech Agent Arena, an improvement from its previous voice agent product, Cartesia Line. Cartesia Managed Agents demonstrated a Task Success Rate of 91.7%, up from 77.5%. Additionally, the open-weights model Ling 3.1 Flash 1019, an update from Ant.g AI, scored 44 on the v4.3 of the Index. Microsoft's MAI-Image-2.7-Preview also performed well on the AA-Image-Editing v2.1 benchmark, positioning itself behind OpenAI's GPT Image 2.5 Sunburst.

The Intelligence Index v5's shift towards private datasets for evaluation signifies a trend where AI companies are increasingly relying on proprietary data to demonstrate their models' superiority. This move makes direct, apples-to-apples comparisons more challenging for external observers and developers, potentially leading to a more opaque market. As models become more specialized, benchmarks may need to evolve to capture nuanced capabilities beyond general performance.

The performance improvements seen in Cartesia Managed Agents and Ling 3.1 Flash 1019 highlight the rapid pace of development in specialized AI applications, particularly in areas like voice agents and reasoning. The close competition in image editing benchmarks between Microsoft and OpenAI suggests an intensifying race for dominance in generative visual AI, indicating that users can expect more sophisticated and capable tools in this domain.