Meter Reading
By Mathieu Kessler, Kesslernity
AI API list prices and the dates that change your bill. USD per 1M tokens unless stated.
Your $40 Argon bill becomes $80 on 1 Dec.
Reference workload: $40 $80 /month
Google doubles two Gemini prices within five weeks: Gemini 4 Argon goes from $2/$10 to $4/$20 on 1 Dec, and Gemini 3.8 Flash from $0.75/$3.75 to $1.50/$7.50 on 1 Jan 2027. On the reference workload (10M input + 2M output a month), Argon stays cheaper than the $10/$50 flagships but costs more than double grok-4.7 ($32) and Mistral Large 4 ($22).
Do: If you picked Argon on price, rerun that comparison before 1 Dec.
Source, checked 23 Oct 2026
Act before
Dates that change your bill, soonest first. Past dates grey out.
DeepSeek starts billing Search/Fetch tool calls made through Harness, Claude Code or Codex: $0.75 per 1,000.
Do: Count your agent tool calls. 100,000 a month is $75 that used to be free.
Source, checked 23 Oct 2026
Vertex AI bills Gemini 4 completions over 32K tokens even when the request then times out.
Do: Set max output tokens, stream long responses, alert on timeout rate.
Source, checked 23 Oct 2026
OpenAI gpt-5.6-sol promo ($4/$20) can end; it runs "at least through" this date.
Do: Budget 2027 at list price, not promo.
Source, checked 23 Oct 2026
Gemini 4 Argon doubles to $4/$20.
Do: Rerun your model comparison now.
Source, checked 23 Oct 2026
Last day to buy the M365 Copilot Business add-on at $18/user/month instead of $21 (annual commitment, first year only, existing eligible Business plans).
Do: If seats are going in during Q1 anyway, the signature date decides the rate.
Source, checked 23 Oct 2026
Gemini 3.8 Flash doubles to $1.50/$7.50.
Do: Test a cheaper fallback for high-volume Flash jobs.
Source, checked 23 Oct 2026
What moved
Changes already in effect.
DeepSeek rerouted deepseek-v4-flash to V4.1-Flash
Affects Anything calling the deepseek-v4-flash name, which now routes to V4.1-Flash.
Do Run your eval set against the new target this week.
Source, checked 23 Oct 2026
Bedrock cut Claude Sonnet 5.5 cache reads by 50% from 7 Oct
Affects Sonnet workloads on Bedrock with long, repeated prompts or RAG context.
Do Move stable context into the cache, then compare against calling Anthropic directly.
Source, checked 23 Oct 2026
Mistral Large 4 now lists at $1.36/$4.18, cached input $0.14
Affects Anyone who shortlisted Mistral but could not price it against the incumbents.
Do Large 3 lists at $0.50/$1.50, so Large 4 is roughly 172% more on input than the model it follows. Compare both generations rather than assuming the newer one is the cheaper replacement.
Source, checked 24 Oct 2026
OpenAI is winding down self-serve fine-tuning
Affects Teams with fine-tuned models in production.
Do List them, find each shutdown date, plan the migration.
Source, checked 23 Oct 2026
What a real workload costs
Standard tier, no caching. Enter your monthly tokens. Price isn’t quality: treat this as a shortlist, not a ranking.
| Source, checked | ||||
|---|---|---|---|---|
| DeepSeek v4-pro (peak) | 1.32 | 3.96 | US$21.12 | Source, checked 23 Oct 2026 |
| Mistral Large 4 | 1.36 | 4.18 | US$21.96 | Source, checked 24 Oct 2026 |
| xAI grok-4.7 | 2.00 | 6.00 | US$32.00 | Source, checked 24 Oct 2026 |
| Google Gemini 4 Argon (to 30 Nov) | 2.00 | 10.00 | US$40.00 | Source, checked 23 Oct 2026 |
| OpenAI gpt-6-astra | 10.00 | 50.00 | US$200.00 | Source, checked 23 Oct 2026 |
| Anthropic Claude Fable 5.1 | 10.00 | 50.00 | US$200.00 | Source, checked 24 Oct 2026 |
Click a column header to sort. Prices in USD per 1M tokens. Standard tier, no caching.
Want this table for your own volumes, plus the Copilot and cloud meters?
AI Meter Map ($39) is the price book I use.
In AI Meter Map ($39): cloud commitments and break-evens, Copilot meters, caching and batch scenarios, price history, export
The price you’d never guess
Claude Fable 5 and 5.1 share a $10/$50 list price, but cache reads cost $1.00 per 1M on Fable 5 and $0.25 on 5.1. Check which model ID your agents actually call.
Source, checked 24 Oct 2026
Get the next reading
Free, monthly, every price sourced. Next issue: Thursday 26 November. Forward this to whoever signs your AI invoices.
Subscribe freeTell me which price change I missed: mathieu@kesslernity.com. Readers who arrive from search cannot reply to an email they never got.