When AI work buys time

OpenAI says GPT-5.6 Luna is 80% cheaper and GPT-5.6 Terra is 20% cheaper. In the same announcement, it offers GPT-5.6 Sol Fast mode in the API at up to 2.5 times Standard processing speed for twice the Standard price.

That puts a practical choice in front of us. A long run of routine checks may reward lower processing cost; an urgent API task may justify paying more to wait less. The approved evidence gives relative changes, not a complete bill.

A model price cut does not settle the value of a workflow. Repeated review, classification and checking work can make a lower unit cost matter quickly. A blocked urgent task has a different cost: time.

OpenAI frames Sol Fast as an API option for that second case. Its up-to-2.5-times-speed claim and twice-Standard-price trade-off are supplier statements, so they describe the offer rather than a measured outcome for our workflow.

OpenAI says Luna’s price falls by 80% and Terra’s by 20%. It also says lower Luna and Terra prices are reflected in how usage is counted in Codex and ChatGPT Work, without giving the underlying prices, quotas or billing mechanics in the approved pack.

For Sol, OpenAI says Fast mode offers up to 2.5 times the speed of Standard processing at twice the Standard price in the API. The comparison is useful as a trade-off, but the pack does not provide workload definitions, measured latency or benchmark method.

The approved OpenAI thread links to an official article titled “Advancing the price-performance frontier with GPT-5.6”. That title explains the company’s framing: lower prices and faster processing are being presented as related choices.

The linked article itself was not captured as an approved source. We can use the visible title and the thread’s words, but we cannot import its uncaptured body, pricing tables or additional claims.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, and offering a faster option for GPT-5.6 Sol in the API.

OpenAI, X thread

OpenAI published an Artificial Analysis Intelligence Index v4.1 chart in the captured thread. It is relevant because it shows the evidence surface OpenAI chose to support its price-performance framing.

Publication is not independent validation. The approved pack does not establish the chart’s methodology, values or comparisons, so the visual remains OpenAI’s supporting material rather than a benchmark we can treat as settled.

We can begin with the job rather than the product name. If the task is routine, high-volume and tolerant of delay, lower processing cost is the sensible question. If it is an urgent API task, the value of a faster response may outweigh the premium.

This is an editorial decision model, not OpenAI’s disclosed routing policy. The point is to make the cost–time trade-off explicit before treating any headline percentage as a recommendation.

For routine work, OpenAI says Luna and Terra’s lower prices change how usage is counted in Codex and ChatGPT Work. We still need the relevant plan, usage and price details before turning that statement into a forecast.

For urgent API work, Sol Fast is a paid option with a stated speed trade-off. Then there is Auto-review: OpenAI says it is moving to Luna and expects the combination to cost about 10 times less. “Expect” and “about” matter; neither substitutes for realised savings in our own workflow.

OpenAI’s announcement is more useful than a simple price-cut story. It presents lower-cost options for repeated work and a premium fast lane for time-sensitive API tasks, then points to Auto-review as a possible workflow-level saving.

The stronger reading is still cautious. Relative price changes, a supplier-posted chart and an expected cost reduction do not complete a workflow calculation. We need our own usage, timing and price evidence before deciding which lane earns the work.

Efficiency is becoming selectable: save on the work we repeat, or pay to move the work that cannot wait.

Synthesis from OpenAI’s announcement, SRC-001