Model Launches
New AI model releases and what they change for builders.
Running Thinking Machines' Inkling without a datacenter
Thinking Machines Lab shipped Inkling on July 15, 2026: a 975 billion parameter open weight model (41B active, Apache 2.0), plus a 276B/12B active preview called Inkling-Small. Full weights need 600GB of VRAM even at 4-bit. Here's the actual hardware, the llama.cpp commands, the hosted pricing against Claude Sonnet 5, and the failure mode that breaks local models in agent loops before the model itself does.
Kimi K3 benchmarks, pricing, and how to start using it
Moonshot AI shipped Kimi K3 on July 16, 2026, a 2.8 trillion parameter open-weight model that activates just 16 of 896 experts per request and reads a 1,048,576 token context window. It already beats GPT-5.6 Sol Max on SWE Marathon (42.0 vs 39.0) and costs $3 per million input tokens through an API today. Full weights land July 27.