Sonar Vortex integrates with AI coding agents to provide architectural context before code generation and verify output in real time. Internal testing shows 36% lower token consumption and 92% fewer defects compared to traditional CI checks that occur after agent completion.

Thinking Machines released Inkling on July 15, 2026. The model features 66 layers with 256 experts per layer, of which only six activate per token. Most layers access short text windows while a few can access all context. Position encoding uses an older method abandoned by most labs.

Founded by Mira Murati, former CTO of OpenAI, Thinking Machines aims to extend human will and judgment through AI. The company focuses on training strong models, customizable tools, communication interfaces, and publishing research. Inkling is their first model trained from scratch, available under Apache 2.0 license on Hugging Face.

Inkling employs Mixture of Experts architecture with 975 billion parameters, though only 41 billion activate per token. The router selects experts using sigmoid scores, with higher-scoring experts contributing more to output.