AI Token Economics Weekly – AI Providers Updates

Week of August 4-8, 2026

OpenAI

GPT-5.6 Just Got Way Cheaper (and Faster) to Run

OpenAI dropped some serious pricing changes for GPT-5.6, and the numbers are hard to ignore. Luna pricing is down 80%, while Terra pricing has been cut by 20%. That’s a massive shift for anyone tracking API spend on these models.

On top of the price cuts, OpenAI also introduced Fast mode, which replaces the older Priority Processing option. It’s designed to give you quicker response times without the extra overhead that came with the previous setup.

For teams managing token budgets, this update is a win on two fronts: lower per-token costs mean more headroom in your budget, and the new Fast mode gives you a leaner way to prioritize speed when you need it. If you’ve been holding off on scaling up GPT-5.6 usage because of cost concerns, now’s a good time to revisit those calculations.

Google Cloud

Gemini Enterprise Now Lets You Pay Only for What You Actually Use

Google Cloud just made Gemini Enterprise’s pay-as-you-go edition generally available, and it’s a big deal for anyone tired of managing pooled user license quotas. Instead of committing to a set number of licenses, you now pay based on actual feature usage.

This edition comes with built-in monitoring for feature usage plus the ability to set monthly spend limits. That means you get visibility into what’s actually costing you money, and you can put guardrails in place before costs creep up unexpectedly.

For FinOps teams, this is exactly the kind of flexibility that makes budgeting easier. No more guessing how many licenses you’ll need or overpaying for unused seats — you just pay for what gets used, and you can cap spend if things start trending in the wrong direction.

New Tracing Makes It Easier to Spot Where Your AI Budget Is Going

Google Cloud also rolled out expanded tracing for Gemini Enterprise data connectors, adding new spans specifically for tool execution and connector invocation.

This might sound like a technical detail, but it actually matters a lot if you’re trying to figure out where your AI workflow costs are piling up. With more granular tracing, you can see exactly which tool calls or connector invocations are eating up resources.

The practical upside here is better observability. When you can pinpoint inefficiencies in your AI workflows down to the specific tool or connector, it becomes a lot easier to optimize spend and cut out the parts of your pipeline that aren’t pulling their weight.