By J. Wren · Last reviewed 2026-10-13
Every rate movement across the models this site tracks, with the date it was picked up.
A scheduled job has checked these rates once a day since 2026-07-25 (runs are occasionally delayed; between 2026-07-30 and 2026-08-03 a guard held back a large change until it was verified by hand). 4 price changes on 4 models have been confirmed (8 input/output rate movements); the most recent was on 2026-08-22, 63 days before the last check on 2026-10-24.
Each marker is a day on which at least one rate changed (green: all cuts, red: all increases, blue: mixed). The same events are listed in the table below.
| Date | Model | Vendor | Change |
|---|---|---|---|
| 2026-08-22 | GPT-5.6 | OpenAI | Input $5 → $4 ▼ 20% |
| 2026-08-22 | GPT-5.6 | OpenAI | Output $30 → $20 ▼ 33% |
| 2026-08-19 | Gemini 3.6 Flash | Input $1.5 → $0.75 ▼ 50% | |
| 2026-08-19 | Gemini 3.6 Flash | Output $7.5 → $3.75 ▼ 50% | |
| 2026-07-30 | GPT-5.6 Terra | OpenAI | Input $2.5 → $2 ▼ 20% |
| 2026-07-30 | GPT-5.6 Terra | OpenAI | Output $15 → $12 ▼ 20% |
| 2026-07-30 | GPT-5.6 Luna | OpenAI | Input $1 → $0.2 ▼ 80% |
| 2026-07-30 | GPT-5.6 Luna | OpenAI | Output $6 → $1.2 ▼ 80% |
Tracking since 2026-07-25 · last checked 2026-10-24 · 16 models watched: Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5, GPT-5.6, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.4 mini, GPT-5.4 nano, Gemini 3.1 Pro (preview), Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite, OpenAI text-embedding-3-small, OpenAI text-embedding-3-large, Gemini embedding 001.
4 rate movements across OpenAI.
4 rate movements across Google, OpenAI.
No rate movements recorded.
No rate movements recorded so far, as of 2026-10-24.
Vendor pricing pages show what a model costs today. They do not show what it cost last month, or when it changed. That history disappears quietly, which makes it hard to answer ordinary questions: was this model always this cheap, is the tier you budgeted for still the same price, did the provider you left get cheaper since.
An automated job checks the rates for every model listed here once a day and records anything that moved. The log below is what it has seen. It is not a reconstruction — nothing is inferred about prices from before tracking started, because that data is not available to this site.
Rates are US dollars per million tokens, quoted separately for input and output. A change entry shows the old rate, the new rate, and the size of the move as a percentage. Models appear as added when they enter the tracked list, which is not the same as the day the vendor launched them. The monthly summary below the log groups the same events by calendar month.
Two things this log deliberately does not do. It does not track every model in existence — only the ones the calculators here use, which are the current mainstream tiers from Anthropic, OpenAI and Google. And it does not claim to be an official record: the data comes from a public community-maintained dataset that follows vendor pricing pages, so a change can show up a little after the vendor announces it.
Once a day, automatically, although runs are occasionally delayed. The job compares every tracked rate against the last published values and writes an entry for anything that differs. If it finds nothing, nothing is added — a quiet log usually means prices held; the last-checked date at the top of this page shows whether the check is still running.
A rate that moves more than 50% in a single day is treated as a data error rather than a price cut. The job stops and publishes nothing, leaving the previous values in place until the change is confirmed by hand. That makes the log slower to react to genuinely dramatic changes, which is the safer trade.
Not reliably. Rate changes follow model releases and competitive pressure, not a schedule. The log is useful for answering what a model costs now and what it cost before, and for noticing that a tier you budgeted around has moved — not for forecasting.
Price it out: LLM API Cost Calculator · Claude · GPT · Gemini · raw data: pricing.json