GPT-6 Luna and average handle time. Average handle time is the minutes an agent spends on one ticket, start to send. Draft-and-review puts an AI draft in the reply box, with the right answer already pulled in, so the agent edits instead of writing from scratch. This is a support lead's read on the price change, not an engineer's read on the model.
On september 22, 2026, OpenAI introduced lower-cost GPT-6 tiers, sol and Luna, per letsdatascience. That is a secondary report, so re-check the details on OpenAI's own pricing page before you budget against them. The direction is the point: mid-tier prices keep falling.
To see how far they fell, look at the frontier line Luna undercuts. GPT-6 astra was released september 3, 2026 with a roughly 1m-token context, per vorplabs. Luna's output price is a small fraction of astra's. Same vendor, three weeks apart.
#Why does a cheaper model change what you run behind a ticket?
How does AI lower average handle time in customer support?
AI lowers average handle time by drafting the reply before the agent opens it, with the relevant knowledge-base answer already retrieved and inserted. The agent reads, edits, and sends instead of writing from a blank box. That cuts the composing minutes, which are the largest movable part of handle time on most teams.
Until now, most teams reserved AI drafting for the hard 20% of tickets. A complex refund, an escalation, a thread with ten messages. The simple ones got a canned macro, because running a fresh draft on every ticket cost more than it saved.
That cost objection is what Luna removes. A support draft with retrieval context runs a few thousand input tokens and a few hundred output tokens. At Luna's rates that is a tiny fraction of a cent for each ticket, so drafting behind every ticket in a high-volume queue costs next to nothing. At astra's rates the same volume added up to a real line item.
#What does draft-and-review actually do to average handle time?
Break one ticket into parts. The agent reads it, finds the answer, composes the reply, and sends. Composing is usually the biggest slice. If a six-minute ticket spends three minutes being written, a good draft takes that to about one minute. The ticket now closes in roughly four minutes. That is a 33% cut, and tickets per agent per shift move the same way.
The agent is still in charge. Nothing sends on its own. What sits behind every ticket is small and visible:
- Retrieval pulls the matching knowledge-base article and policy into context
- The model drafts a reply in your tone, citing the source it used
- The agent edits the draft and presses send, or rewrites it if it is wrong
- The correction feeds back so the next draft on that topic is better
This is the same shape of work as a worked example of agent-side retrieval for a support team, an illustrative setup with assumed numbers, not a client result. The idea there was not a smarter model. It was putting the answer in the agent's hands before they went looking. Luna's price just lets you do that on the whole queue instead of a slice of it. We design these rollouts through our AI consulting practice and build the retrieval layer through AI implementation.
We are
Stennir puts a draft in your agent's reply box so a human edits and sends faster.
We aren't
Stennir is not a customer-facing deflection bot that answers your users instead of your team.
#Should you put AI behind every ticket now?
Mostly yes, with two checks. First, re-verify Luna's price on OpenAI's pricing page, because the pricing we cite is secondary reporting from september. Second, gate on quality, not cost. A cheap draft that is wrong half the time raises handle time, because the agent rewrites it. Measure the edit rate before and after. Keep the hard tickets on a stronger tier if the drafts there need it.
The old question was which tickets can we afford to draft. Luna changes it to which tickets should a human still write by hand. That is a much shorter list.
If you lead a support team and you track handle time and tickets per agent per shift, the move this quarter is to price a draft-and-review pilot on your real volume and edit rate, not on a vendor's demo. We are new, and the terms we quote are the ones we build to. book a 30-min discovery call and we will size it against your own numbers before you commit to anything.
