Researchers at NVIDIA, NTU, and MIT developed SoL-Pi, a system that optimizes coding agents through recursive auto-research loops across diverse environments. SoL-Pi reduces token usage by 44.7-49.0% and API costs by approximately one third while maintaining performance comparable to existing systems on the 51-task EdgeBench evaluation.
Grok 4.7 is launching soon with 2.1 trillion parameters as a new pretraining effort rather than a refresh of 4.6. The model outperforms 4.6 in most areas with improved token efficiency and benefits from SpaceX and Starlink engineering data, though multimodal capabilities need further development and serving speed is slightly slower.
Novgraph is a persistent knowledge graph tool for codebases designed to reduce token usage by 96-99%. It offers features including recorded intent, ranked blast radius, architecture context, precise code discovery, and continuous history tracking.
An article categorizing AI model inference calls into four functional roles: Initialization (processing task-independent context), Orchestration, Reasoning (resolving task-relevant uncertainty), and Synthesis. The taxonomy distinguishes inference's functional purpose beyond aggregate token metrics, enabling more precise measurement of inference value versus waste in agent systems.