CHANGELOG

What Shipped This Week: GPT-5.6 Luna on Every Plan

Every agent on CloudAxis now runs on GPT-5.6 Luna by default — Free through Max, no dropdown, no API keys. We also shipped a full week of CloudAxis Coding polish and fixed the chat onboarding path that was losing new users.

5–7 min read

Open Chat on a Free plan this morning and you get the same model as Pro: GPT-5.6 Luna, routed automatically, capped by your plan credits. No picker. No guessing which model your Research agent is burning.

Under that platform shift, we kept tightening CloudAxis Coding — stop button, live preview links, branch picker — and fixed the empty-chat path that was sending new signups into a dead-end welcome screen.

Auto Mode is GPT-5.6 Luna on every plan

Before this week, Auto Mode on CloudAxis meant DeepSeek Flash. Pro and Max users could open a model picker and choose GPT, Claude, or Kimi manually. Free and Growth users could not. That split created two problems: new users on Free got a different agent brain than paying teams, and anyone who picked the wrong model could burn credits faster than Auto would have.

We removed the picker entirely. Every plan — Free, Growth, Pro, Max — now runs Auto Mode on GPT-5.6 Luna. Open Settings and you will see your plan cap and credit balance, not a dropdown of model names. Specialists, Cloudia, and scheduled duties all inherit the same default. Your Research agent at 6:47am and your Content agent at 3pm run on the same hosted model, billed against the same credit pool.

Why Luna specifically: OpenAI cut GPT-5.6 Luna pricing by 80% on July 30. At $0.20 per million input tokens and $1.20 per million output tokens, Luna sits in the sweet spot for agentic work — long context, reliable tool calling, predictable cost per duty run. We measured Luna on our own agent harness (inventory policy → reorder confirmation) and published the numbers in our agentic harness benchmark. Auto Mode is how we pass that economics to you without making you read a pricing table.

The practical effect on your daily workflow: a Growth plan running three morning duties — competitor scan, inbox triage, weekly report draft — now burns credits at Luna rates instead of whatever model someone picked last Tuesday. Hard monthly caps still apply. You still know your maximum spend. What changed is consistency: every agent, every plan, same brain, same billing math.

We also fixed a backend compatibility issue where Luna with tools could inherit a reasoning-effort setting that slowed multi-step agent runs. Tool-heavy duties — browser automation, code execution, file writes — should feel snappier now. If you had agents that seemed to stall between tool calls after the migration, that fix is live.

Also shipped

Smaller changes that landed in the same release window:

What's next

CloudAxis Coding is still in beta. More language toolchains (Go, Rust, Ruby, PHP) and monorepo detection are next on the build-machine list. On the model side, Auto Mode stays Luna-first — we will announce routing changes before they ship, not after.

Frequently asked questions

Can I still pick a model manually?

No. Auto Mode on GPT-5.6 Luna is the only option on every plan. Your plan cap is your absolute maximum spend.

Did my existing agents change models automatically?

Yes. All specialists and Cloudia now run on Luna through Auto Mode. No action required on your side.

Related reading in this series
GPT-5.6 Luna complete guide · What it costs to run AI agents · Last week: CloudAxis Coding beta