AI News: OpenAI Price Cuts, Inkling-Small, and Gemini Robotics 2

OpenAI implemented significant price cuts for its GPT-5.6 models, reducing Luna by 80 percent and Terra by 20 percent while introducing a faster Sol tier. These cost reductions heavily impact agent workflows, enabling tools like the ChatGPT app auto-review and Codex CLI to operate at roughly ten times lower cost. Meanwhile, technical discussions around the ARC-AGI-3 benchmark highlighted that modern evaluations test entire agent systems—including memory retention, context compaction, and tool orchestration—rather than just base model weights.

Thinking Machines released Inkling-Small, an open-weights, natively multimodal Mixture of Experts model featuring 276 billion total parameters and 12 billion active parameters. The model supports audio, images, and text jointly alongside a one-million-token context window. In the robotics sector, Google DeepMind launched Gemini Robotics 2 for full-body control and multi-robot coordination, featuring an embodied reasoning model that handles motor tasks and recovers from failed steps.

Cloud development agents achieved major production milestones, with Cursor reporting that 56 percent of merged pull requests now originate from cloud agents equipped with dedicated cloud computers. Additionally, local AI communities focused heavily on extreme quantization and optimization of massive open models like Kimi K3, running the 2.8-trillion-parameter model locally on consumer hardware and Apple Silicon despite low initial throughput.