AskMeAnything
Ask
Ask
38
@scope_of_work_sky ·

OpenAI cut GPT-5.6 Luna by about 80% on 30 July: should Luna be my default now, or is that a downgrade? Pricing

The announcement on 30 July cut API prices on the GPT-5.6 line: Luna by roughly 80%, to about $0.20 per million input tokens and $1.20 per million output, and Terra by around 20%, to about $2 and $12. Sol's standard pricing was left alone. The stated reason was serving efficiency improvements rather than any change to the models themselves. There is also a Fast mode on the API that is substantially quicker and costs more per token.

My pipeline sends everything to the middle tier out of habit, because that is where I landed when the family launched. At the new prices the cheap tier is close to free for my volume, which is about 40 million input tokens a month of document extraction and classification, plus a small amount of agentic work with tools.

What I do not want is to find out over a fortnight that the cheap tier is subtly worse at the one field that matters, in a way I only notice when a customer does. For those who have moved workloads down a tier since the cut: which kinds of work survived the move and which came back?

7 answers Share
Report

7 answers

BestTopNewOldControversial

Answering anonymously — a moderator will review it first.

  • @maren_dowd · 2h ago · 2 replies
    @maren_dowd ·

    Route by task shape, not by price, and move one workload at a time.

    What has moved down a tier cleanly for me: schema-constrained extraction where the schema does the work, classification into a small label set, format rewriting, summarising a document that is already in context. Anything where the answer is mostly transcription under constraints.

    What came back up: multi-step work with tools, where a cheap model's small planning errors compound across calls, and anything requiring it to notice that the document does not contain the answer. Abstention is consistently the first thing to go.

    With 40M input tokens a month you are the ideal candidate for the move, but do it behind an eval and one workload at a time, so when quality dips you know which change caused it.

    21
    Share
    Reply

    Answering anonymously — a moderator will review it first.

    Report
    • @proof_pilar · 2h ago
      @proof_pilar ·

      And put that eval in CI with the model name as config. Then the next price change is a one-line diff and a green or red run, instead of a fortnight of arguing about whether it feels worse.

      9
      Share
      Reply

      Answering anonymously — a moderator will review it first.

      Report
  • @petra_lindqvist · 2h ago
    @petra_lindqvist ·

    The specific danger in extraction pipelines is not wrong answers, it is emptier answers. A cheaper model tends to leave optional fields null rather than infer them, on some small percentage of rows, and every row still validates.

    So before you switch, add two per-batch metrics: null rate per field, and schema-validity rate. Record a fortnight of baseline on your current tier. Then a downgrade that costs you 3% of a field shows up as a number the day it happens rather than as a support ticket in six weeks. Cheap to build, and it pays for itself the first time a model changes underneath you without a version bump.

    14
    Share
    Reply

    Answering anonymously — a moderator will review it first.

    Report
  • @porosity_nerd · 2h ago · 2 replies
    @porosity_nerd ·

    A cut that size means they quantized it harder. You are being offered the same model with the quality shaved off, and the price is the honest signal of what you are now getting.

    10
    Share
    Reply

    Answering anonymously — a moderator will review it first.

    Report
    • @maren_dowd · 2h ago
      @maren_dowd ·

      That is an assumption, and the announcement says otherwise: the stated cause was efficiency improvements across the training and inference stack, and the models on offer and their identifiers did not change. Serving cost falling over time is the normal shape of this industry, not evidence of a nerf.

      Which is not the same as saying quality is guaranteed stable. Providers do change what sits behind a name. The correct response is the same either way: measure it on your own data and keep measuring, rather than inferring quality from the price in either direction.

      12
      Share
      Reply

      Answering anonymously — a moderator will review it first.

      Report
  • @yield_on_cost · 2h ago
    @yield_on_cost ·

    At roughly $0.20 per million input, prompt size stops being the thing that hurts you. Before you rewrite prompts to save tokens, look at what your bill is actually made of now: for most people at these prices it is output tokens, retries, and any fast/priority mode they enabled once and forgot.

    I spent two days last quarter compressing prompts for a workload where input was 6% of the cost. Check the composition first.

    8
    Share
    Reply

    Answering anonymously — a moderator will review it first.

    Report
  • @abd_avery · 2h ago
    @abd_avery ·

    One clarification for people arriving from a chat product: this was API list pricing. If you use the model through a subscription rather than by key, none of these numbers are the ones you are billed on, and your allowance did not necessarily change on 30 July. Different page, different rules.

    5
    Share
    Reply

    Answering anonymously — a moderator will review it first.

    Report