AI Pulse · Frontier Labs Watch

OpenAI just dropped prices on its GPT-5.6 models, and for the first time a frontier lab has said the quiet part out loud. The Luna tier fell 80 percent, now $0.20 per million input tokens and $1.20 output. Terra dropped 20 percent. Sol held its price but got 2.5 times faster. The company named Uber and Microsoft explicitly, saying enterprise customers were blowing their budgets on agentic coding tasks and pointing to Claude Code as a specific cost pain point, according to CNBC.

This matters because it confirms what finance teams have been whispering for months. Agentic workflows, especially code generation running in loops with tool calls and retries, rack up token counts fast. When an agent rewrites a function six times or pulls context from a dozen files, you are not paying for one answer anymore. You are paying for dozens of calls. OpenAI admitting that two of the largest enterprises in the world hit budget ceilings is a public acknowledgment that the current pricing model does not fit how agents actually work in production.

It also puts immediate pressure on Anthropic. If GPT-5.6 Luna now undercuts Claude on price while OpenAI calls out Claude Code by name as a budget problem, every procurement team with a Claude contract has fresh leverage to renegotiate. Expect pricing conversations across the board to reset in the next quarter.

For operators, this is a reminder that agentic AI is not just a technical decision. It is a budget decision. If you are running agents in production or planning to, you need visibility into token spend by task, by agent, by workflow. That is the kind of real-time cost tracking we build into the Omni Command Centre, because guessing at your AI spend in arrears is how you end up in the same room as Uber and Microsoft.