Blog · November 5, 2026 · 4 min read
Four LLM Price Changes by January 1. Budget the Usage, Not the Rate Card.
Between October 26 and January 1, four of the list prices a small business budgets against move on announced dates, and two of them double on the same morning. OpenAI's new API credits expire on a fifth date, your billing day, and no provider emails you before they run out. A 2027 AI budget pinned to this month's rate card is wrong by a date you can look up. The one line that only moves in one direction is usage, so that is the line to budget and to alert on.
The LLM price changes already on the calendar
The LLM price calendar, October 26 to January 1
List price per million tokens, input / output, as published by each vendor
- Oct 26GPT-6.1 TerracutLaunches at $1 / $5, which OpenAI calls 50% below GPT-5.6 Terra
- Nov 30GPT-5.6 Solup$4 / $20 is promotional until November 30; the next price is not published
- Dec 1Gemini 4 Argonup$2 / $10 becomes $4 / $20; cached input $0.10 becomes $0.20
- Jan 1Gemini 3.8 Flash, 3.6 Flashup$0.75 / $3.75 becomes $1.50 / $7.50
- Billing dayOpenAI API creditsreset$100 to $500 a month, spent first, expire with the cycle, nothing rolls over
Three of the five rows are increases. Gemini is not a PulseMeter provider yet; the prices are here because they are on the calendar, not because we watch them.
Why cheaper tokens still produced a bigger bill
CloudZero's October AI Economics Pulse puts AI at 3.02% of the median organization's cloud bill in September, up from 2.66% in August, the fourteenth straight monthly rise and 4.5 times a year earlier. 68% of its 422-organization panel spent at least $1,000 a month on AI. CloudZero's own summary: "The token got cheaper. The bill didn't." The panel is enterprise cloud customers, so take it as direction, not as a small-business benchmark.
The cleanest single number is at the API layer. JPMorgan's October data-center note, as reported by TechFlow, tracked OpenRouter traffic: token volume rose 25% month over month, the simple average price per token fell 5%, and spending rose 40%. Spend shifted toward pricier models, so the usage-weighted price rose 12% while the sticker average fell. OpenRouter is developers, startups and coding agents, not the whole market.
Cheaper tokens, bigger bill: OpenRouter traffic, October 2026
Month-over-month change, JPMorgan as reported by TechFlow
Citadel Securities' "Tokenomics 2.0" note of October 28 is the counterweight. Consumption-weighted token prices fell about 50% over six months and volumes rose about five times, yet Ramp's card and invoice data shows essentially no growth in overall enterprise AI spend in August and September. The developer-and-agent layer accelerated while the broad enterprise bill went flat. Which way your bill moves depends on which layer you buy from. The thing you control is usage.
Finance plans to spend more either way. CloudZero's survey of 252 senior finance leaders found 87% raising FY2027 AI budgets, 51% over their FY2026 plan, and only 12% able to break AI spend down finely enough to act within the quarter. CloudZero's advice is worth repeating: "Budget for usage, not price. Price cuts get spent as volume."
The credit cliff nobody emails you about
Since October 28, ChatGPT Pro subscriptions carry $100, $200 or $500 of API credits per monthly billing period, and Business workspaces get $20 per Standard seat and $100 per Premium seat, capped at $500 a month. Credits are spent before purchased credits, expire at the end of each billing cycle, and never roll over. When they are gone, requests fail with HTTP 429 and the error code credit_balance_exhausted unless auto-recharge is on. OpenAI's own wording: a reported API cost "is not necessarily the cash charged after credits are applied."
So a workload that fits inside $500 of free credit either goes dark on the day the credit runs out or starts billing the card. Either way the date that matters is the exhaustion date, and nobody sends a warning before it.
Three things to do in November
Reprice December from October's volume, not October's bill. Pull token counts by model for October and multiply by the December 1 prices in the table. If Argon or Sol carries a third of your tokens, that line doubles while everything else stays flat.
Compute your credit exhaustion date. Divide the monthly credit by your daily API run-rate. $500 of credit at $40 a day runs out on day 13. Decide now whether auto-recharge is on, so the card bills, or off, so the job stops, and tell the person who owns the key.
Set a run-rate alert that fires inside the month. A month-end budget line tells you what happened. A run-rate alert on each key, routed to the key's owner, tells you on the day that spend is on pace to pass the budget, before the invoice does.
What PulseMeter shows, and what it does not
PulseMeter reads provider usage through read-only keys and shows list-price spend before promotional credits, reconciled nightly to the provider's own cost report. Our OpenAI connect screen says the same thing: the cash on your invoice can be lower than the spend we show, and the alert worth having is the one that fires before the credits run out. Gemini is on the roadmap, not in the product.
Budget the run-rate, not the rate card. PulseMeter watches it hourly on paid plans and tells a human before the month does.
Sources: OpenAI API pricing (GPT-6.1 Terra $1 / $5 / $0.05 cached; "GPT-5.6 Sol's promotional pricing is available until November 30, 2026"; fetched 2026-11-05) and Introducing GPT-6.1 Terra (2026-10-26, the "50% lower" claim). Gemini API pricing (Argon $2 / $10 through November 30 and $4 / $20 from December 1, 2026; Gemini 3.8 Flash and 3.6 Flash $0.75 / $3.75 through December 31 and $1.50 / $7.50 from January 1, 2027; fetched 2026-11-05). OpenAI Help Center, API credits benefit (tiers, expiry, HTTP 429 credit_balance_exhausted; fetched 2026-11-05). CloudZero, AI Economics Pulse, October 2026 (published 2026-10-13, updated 2026-10-29; 422 organizations, enterprise cloud customers). CloudZero, AI budget season (2026-11-02; survey of 252 senior finance leaders dated 2026-10-07). TechFlow, compiling JPMorgan's data-center report (report dated 2026-10-26; we have not read JPMorgan's report itself; OpenRouter traffic only). Citadel Securities, "Tokenomics 2.0" (Frank Flight, 2026-10-28; market commentary labeled not investment research; Ramp's AI spend covers subscriptions, coding agents, API tokens and GPU cloud together).