OpenAI cut GPT-5.6 Luna's price 80% — what it actually means for your automation bill

What happened

On July 30, 2026, OpenAI cut the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, down from $1 and $6. Terra dropped from $2.50/$15 to $2/$12 per million tokens. Sol, the flagship model, kept its price but gained a Fast mode that runs about 2.5x faster for double the cost. OpenAI attributed the cuts to efficiency gains in the model and its serving infrastructure, saying each generation should "accomplish more work at a lower cost."

The timing isn't an accident. It landed days after Anthropic shipped Claude Opus 5 at the same price as its predecessor, and after Google rolled out Gemini 3.6 Flash and 3.5 Flash-Lite, both built around cheaper inference. Three labs, three moves toward lower cost per token, inside the same two weeks.

What's actually new here

An 80% cut on Luna is a real number, not a rounding adjustment. At $0.20 per million input tokens, Luna is now priced closer to a small open-weight model than to a frontier one, while still carrying a 1.1M-token context window and 128K max output. That combination — long context, cheap tokens — is the part worth noticing. A year ago you paid a premium for either speed or context length. Now you can get both on the cheap tier and reserve Sol for the fraction of calls that actually need top-end reasoning.

The competitive read matters too. When three vendors move the same direction in the same stretch, it means the price floor for "good enough" reasoning is dropping across the board, not just at OpenAI. That's the number to track, more than any single vendor's list price.

What it means if you're running automation on this stuff

If you have a pipeline that classifies support tickets, extracts fields from documents, drafts first-pass replies, or routes requests, and it's currently pointed at GPT-5.6 Luna or a comparable mid-tier model, your per-call cost just dropped by a factor of five on the input side. For anything high-volume — thousands of calls a day — that's real money back, not a rounding error on the invoice.

It also changes the build calculus. Tasks that previously got routed to a pricier model because the cheap tier's context window was too short no longer have that excuse. Luna's 1.1M tokens covers most contract review, long email threads, or multi-document comparisons that used to force an upgrade to a flagship model just for room to fit the input.

Reportedly, companies like Uber and Microsoft have been burning through AI budgets faster than planned. Cuts like this are a direct response to that pressure, and a business running its own automation should expect this pattern — vendors competing on cost per token — to continue for a while.

The honest caveat

A cheaper token is not a more reliable one. Nothing about this price cut touches accuracy, hallucination rate, or how often the model needs a retry to get a task right. If your workflow currently needs three calls to Luna to get a clean extraction — one first pass and two corrections — an 80% price cut on each call still leaves you paying for three calls. The sticker price dropped; your actual cost per completed task might not have dropped by anywhere near as much.

There's also a tier trap here. The discount applies to Luna and Terra, the mid and lower tiers. If your task genuinely needs Sol-level reasoning, none of this helps you, and Sol's price didn't move.

What to do about it

Don't just swap the model name in your config and assume the savings show up. Pull your existing eval set — the test cases you already use to check output quality — and rerun it against Luna at the new price point before moving any production traffic. If accuracy holds, you've found real savings. If it doesn't, the discount doesn't apply to you yet, no matter what the price sheet says.