Originally published on rohitraj.tech
Anthropic shipped Claude Opus 5.5 and OpenAI shipped GPT-6 Sol ninety minutes later, both at roughly half the old price. The same 48 hours also put a 27B multimodal model into 5.95GB and taught transformers to run llama.cpp's Metal kernels in-process. The cost floor moved at both ends of the stack at once — here is what actually changed and what to do about it.
Read the full version with code samples, diagrams, and architecture details: This Week in AI Dev: Frontier Prices Halved and a 27B Model Fit in 6GB (Week 39 of 2026)
Work with me: MCP Integration Consultant · Claude Code Consultant · 6-Week MVP Sprint
More engineering notes: rohitraj.tech/notes
Top comments (0)