Thinking Machines Lab released Inkling-Small, a 276B-parameter mixture-of-experts model with 12B active parameters that delivers the same multimodal reasoning capabilities as its larger predecessor while using substantially less compute.

The model retains Inkling's variable thinking effort, 1M-token context window, and multimodal reasoning across text, images, and video. The smaller footprint makes advanced AI capabilities more accessible to organizations with limited computational resources.

OpenAI slashes GPT-5.6 pricing

OpenAI cut GPT-5.6 Luna pricing by 80% and Terra pricing by 20%, while improving Sol's API speed. The price reductions extend across API usage, Codex, and ChatGPT Work subscriptions.

The efficiency gains reflect OpenAI's ongoing optimization of its inference infrastructure. The company has been under pressure to reduce costs as competition intensifies from open-weight models that offer similar performance at lower prices.

Cursor detailed how it improved cloud development environments for AI agents, helping them grow from authoring 10% of merged pull requests to more than half. The company focused on making environments easier for agents to understand, run, and test code.

Google launched Gemini Robotics ER 2, enhancing automation capabilities with advanced AI and LLM integration. The system targets industries requiring precise automation and improved robotic efficiency.

Open models reach parity

Open-weight models have achieved accuracy parity with closed models in regulatory and clinical tasks while delivering significantly lower costs. The ClinReg benchmark showed models like GLM 5.2 and Moonshot AI's Kimi K3 performing within one standard deviation of GPT-5.6 Sol at one-third the cost.

MiniMax unveiled H3, an open multimodal model that generates up to 15 seconds of 2K video with native stereo sound. The model handles unified context across text, images, video, and audio while excelling at instruction following and accurate text rendering.

China's sovereign AI strategy shifted when Moonshot AI open-sourced Kimi K3 on July 27. Any government, company, or individual can now run and retrain the high-quality model for free, reducing ongoing licensing costs and increasing hardware investment returns.