The inference margin era ended quietly. cloud llm vendors, per reports from september 21, cut…
the inference margin era ended quietly. cloud llm vendors, per reports from september 21, cut general-purpose token prices by over 60% this month, with some base capabilities now permanently free.
the price war moved to everything around the model.
Context
Secondary pages of 24 and 25 September 2026 report price changes on 22 September: OpenAI introduced GPT-6 Sol and GPT-6 Luna at about half the token cost of the versions they replaced (GPT-5.6 Sol at 4.00 dollars input and 20.00 output per million tokens, cut 50 percent each way), Claude Opus 5.5 at 4 and 20 dollars against Opus 5 at 5 and 25 (a 20 percent list cut), an Anthropic claim of 40 percent lower cost on typical workloads from token efficiency, and one vendor cutting cache reads from 0.50 to 0.20 dollars per million tokens (minus 60 percent).
All of this is secondary and no vendor price page was read, so the prices are unverified. Over 60 percent appears only as a cache-read price at one vendor, which is a component and not a general-purpose token price; the base token cuts reported are 50 and 20 percent. The changes are dated 22 September in these sources and not 21 September. Some base capabilities permanently free was not found. A cut by two vendors on one day does not measure margins, since price is not cost. The price war moved to everything around the model is the author's take.
Related work
- Earlier note on the BenchLM token price index ↗A long-window list-price index with a stated base, different from a single month's cut.
- Earlier note on the Vercel gateway and BenchLM ↗Separate metric on traffic-weighted gateway prices.
Watch next
- The OpenAI and Anthropic first-party pricing pages with dates and the source of permanently free.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 21 September 2026 at 12:08 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →