OSASI.ORGCommentary · Ideas for our profession
All briefs← Back to OSASI

$600 to $7,000 per day: who is paying for AI?

September 24, 2026 · OSASI Commentary

The show's message is that cost is the story: OpenAI's new models halve API prices, and every drop in price opens up more work for AI. OpenAI's own figure shows where that leads: valued at API prices, its median researcher uses more than $600 of tokens a day, and $7,000 at the 90th percentile.

How a brief is made

A member reads an article or listens to a podcast and sends the snippets, key thoughts or concerns tied to a passage or a timestamp. An AI assistant, chosen for the task and the setup (Claude, ChatGPT, Codex, Gemini or another), searches the full article or transcript, pins the times, checks the quotations, and assembles a draft. The draft goes through rounds of review, and nothing is published until a person approves that exact version. The cues on each post start the player at the passage discussed. Commentary is OSASI's perspective, offered for discussion, not a summary of the source. Each brief is co-authored by a person and an AI assistant. AI assistant for this brief: Claude (Anthropic).

Source podcast · The AI Daily Brief

This episode prompted the commentary below. It is one perspective, offered for discussion.

The AI Daily Brief · September 23, 2026
Opus 5.5 vs GPT-6 Sol and Luna

Listening highlights · Nathaniel Whittemore, host, on his own. The capture that prompted this brief is at 8:19.

−31:02 to −29:09 remainingIntroductions · Nathaniel WhittemoreWho is speaking, four model releases in two days, and why "which is better" is the wrong question
−25:12 to −23:35 remainingSol and Luna: all about costAPI prices down 50 percent; OpenAI charts performance against cost instead of benchmark tables
−23:35 to −22:24 remainingTons of AIPer-task pricing, and OpenAI's figures: $600 a day in tokens for its median researcher, $7,000 at the 90th percentile
−8:43 to −7:41 remainingThe best model at every price pointOpenAI's strategy: a model slate for an era of many models, not one

Tap a time above to start the player there. If Spotify shows a card about the show first, close it and the player carries on at that time. Times are approximate transcript times; ads and episode editions may shift them. This plays the full episode, not an extracted clip.

Who is talking

Nathaniel Whittemore hosts The AI Daily Brief, a daily show on AI news and analysis. This episode has no guest: he walks through the same-day releases of Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna, reading from the announcements and from posts by people testing the models, among them Sam Altman of OpenAI.

The show's own message

Cost is the story of this release. The Sol and Luna announcement, Whittemore says, "was all about cost and efficiency", with API prices half those of the previous generation, and in the show's telling every drop in price opens up more work that AI can take on.

Tons of AI

Altman, as read on the show, put the aim plainly: "We want people to be able to use tons of AI." OpenAI's announcement put a size on "tons":

"Valued at API prices, daily token usage has exceeded $600 for the median researcher and $7,000 for researchers at the 90th percentile"

Two words in that sentence carry the weight: "valued at". That is what the tokens would cost a customer at list price, used freely by the people building the models. At about 250 working days a year it comes to roughly $150,000 a head at the median and $1.75 million at the 90th percentile. And the price of those tokens has just been cut in half, with Altman saying OpenAI wants its API to offer "the best model at every price point". On the same day, Sanchit Vir Gogia of Greyhound Research, quoted by InfoWorld, described the pricing as "a land grab".

February to September

The price cut has a back story. In February, The Information reported that both labs had missed their own gross-margin forecasts for 2025 because running the models cost more than planned. By June it was reporting that OpenAI engineers had found a way to cut those costs in half, and this week OpenAI said as much: "Improvements in caching and inference let us serve these models at lower cost, and we're passing those savings directly on to users and customers."

So part of the cut is real efficiency. Whether the business as a whole pays for itself is a separate question, and here the two labs now diverge. Anthropic reported a positive adjusted operating result for its second quarter and expects another this quarter. OpenAI's own projections, as reported, imply a loss of about $14 billion in 2026.

What this means for OSASI

A profession should not build its core work on a price set during a land grab. The answer is not to avoid AI; it is to use it where it earns its keep and to depend on it less where the work must be repeatable.

For OSASI that points to building out reproducible code tools that rely less on generative AI, for a more sustainable system: software that gives the same answer from the same inputs every time, can be checked line by line, and costs the same to run next year as it does today. Generative AI can help write and test those tools. It does not need to be in the loop every time a report is produced. Advanced AI is best reserved for the edge cases and the nuanced analysis where it adds the most. An appraisal that rests on reproducible code can be explained and defended long after the pricing of any model has changed.

Further reading

OpenAI: Introducing GPT-6 Sol and Luna, the source of the $600 and $7,000 figures and the pricing table.

The Information: Why OpenAI, Anthropic missed their own gross margin forecasts (subscription). Reported OpenAI's 2025 adjusted gross margin at 33 percent against a 46 percent forecast.

The Information: "Tokenmaxxing" and the AI margin squeeze (subscription, April 2026), on heavier agent-style usage and what it does to costs.

Fortune on Anthropic's valuation (August 2026): a company that has "barely turned a profit" and what its valuation would require.

Epoch AI, quoted on the show: at a given level of performance, AI cost has fallen about 47 percent a quarter since 2023.

SemiAnalysis's subscription test, as reported by ScienceBlog: a $200-a-month ChatGPT Pro plan pushed to its limits could use as much as $14,000 a month of tokens at API prices. The original report is paywalled, and secondary accounts differ on how the Claude plans were priced, so the Claude figures are left out here.

Send a thought →

A like is counted once per reader per day and stores nothing about you. A thought goes by email to the person who edits these briefs; nothing you write appears on the site unless it becomes part of an approved brief. Share opens the service you pick; this site adds no trackers.

Prefer a feed reader, Slack or Teams? Follow by RSS.