Price changes, past and scheduled

Providers almost never announce a price rise loudly. It arrives as a footnote, as an introductory rate quietly expiring, or as a billing restructure. This page collects both what has already changed and what is already scheduled, so a cost model built today does not silently go stale.

Already scheduled

2027-01-01 Google increase
Gemini 3.6 and 3.7 Flash prices double
The Google pricing page states current rates are valid only through 2026-12-31 and double on 2027-01-01: input rises from $0.75 to $1.50, output from $3.75 to $7.50, and context-cache reads from $0.075 to $0.15 per million tokens. Caching does not soften the increase — every rate doubles together. Any cost model built on current Flash pricing needs revisiting before year end.
Gemini 3.7 Flash · Gemini 3.6 Flash · official source

Already in effect

2026-10-05 DeepSeek decrease
DeepSeek retired V4 Flash; V4.1 Flash bills less, especially on cache hits
DeepSeek now lists deepseek-flash (DeepSeek-V4.1-Flash) as its Flash model. The legacy name deepseek-v4-flash is still accepted but is served by V4.1 Flash at the new rates. At peak, input fell from $0.44 to $0.30, output from $1.32 to $1.20, and cache hits from $0.014 to $0.006 per million tokens; off-peak rates remain half of peak ($0.15, $0.60 and $0.003). Peak hours are unchanged (01:00-04:00 and 06:00-10:00 UTC, Monday through Friday) but now exclude Chinese public holidays, which bill off-peak in full. V4 Pro pricing is unchanged. DeepSeek published no effective date; the change was observed between the 2026-08-24 and 2026-10-05 verification passes.
DeepSeek Flash (V4.1) · official source
2026-08-31 Anthropic decrease
Claude Sonnet 5 introductory $2/$10 became the standard price
The $2/$10 rate was announced as introductory pricing through 2026-08-31, with a rise to $3/$15 scheduled for 2026-09-01. Anthropic has since confirmed that increase will not happen and $2/$10 is now standard. Teams that budgeted for $3/$15 can revise downward.
Claude Sonnet 5 · official source
2026-08-24 OpenAI restructure
GPT-5.6 Sol got cheaper, but the whole 5.6 family gained a 272K price cliff
Sol dropped from $5.00 to $4.00 input and $30.00 to $20.00 output per million tokens, with cached input falling from $0.50 to $0.40. At the same time OpenAI split the 5.6 family into "short context" and "long context" columns divided at 272,000 tokens. Above that threshold Sol bills $8.00 input and $30.00 output, Terra $4.00 and $18.00, and Luna $0.40 and $1.80 — input doubles and output rises 1.5x. For prompts under 272K this is a straightforward 33% cut on Sol output; for prompts above it, Sol output is unchanged at $30.00 and input is now 1.6x what it used to be. OpenAI published no effective date; the change was observed between the 2026-08-18 and 2026-08-24 verification passes.
GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · official source
2026-08-16 DeepSeek restructure
DeepSeek V4 moved to time-of-day pricing and raised rates sharply
V4 Flash output went from a flat $0.28 per million tokens to $1.32 at peak and $0.66 off-peak. V4 Pro output went from $0.87 to $3.96 peak and $1.98 off-peak. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday; every other hour — including the entire weekend — bills at half the peak rate. Measured at peak, output pricing rose 4.7x; even off-peak it rose 1.4x.
DeepSeek Flash (V4.1) · DeepSeek V4 Pro · official source

Retirements are repricing events too

When a model retires you are moved onto its replacement, which is usually priced differently. A shutdown date is therefore a scheduled repricing event for anyone still on the old model — and unlike a price change, it breaks outright rather than costing more. The next confirmed shutdowns:

Date Model Provider Replacement
2026-10-23 gpt-3.5-turbo-0125 OpenAI gpt-5.6-terra
2026-10-23 gpt-4-0613 OpenAI gpt-5.6-sol
2026-10-23 gpt-4-1106-preview OpenAI gpt-5.6-sol
2026-10-23 gpt-4-turbo OpenAI gpt-5.6-sol
2026-10-23 gpt-4.1-nano OpenAI gpt-5.6-luna
2026-10-23 gpt-4o-2024-05-13 OpenAI gpt-5.6-sol
2026-10-23 o1-2024-12-17 OpenAI gpt-5.6-sol
2026-10-23 o1-pro-2025-03-19 OpenAI gpt-5.6-sol
2026-10-23 o3-mini-2025-01-31 OpenAI gpt-5.6-sol
2026-10-23 o4-mini-2025-04-16 OpenAI gpt-5.6-terra
2026-10-23 gpt-image-1 OpenAI gpt-image-2
2026-11-30 claude-sonnet-4-5-20250929 Anthropic claude-sonnet-5-5

All 73 announced retirements with live countdowns →

Get told when this changes

Subscribe to /feed.xml in any RSS reader. It carries every price change and every confirmed model shutdown as it is recorded — no account, no email, nothing to unsubscribe from.

A price rise you find out about from your invoice has already cost you a month.

Frequently asked questions

Which LLM API prices are going up?

Gemini 3.6 Flash and Gemini 3.7 Flash double on 1 January 2027 — input goes from $0.75 to $1.50 and output from $3.75 to $7.50 per million tokens, per a footnote on Google’s own pricing page stating current rates are valid only through 31 December 2026. DeepSeek already raised V4 sharply on 16 August 2026 when it moved to peak/off-peak billing: V4 Flash output went from a flat $0.28 to $1.32 at peak, a 4.7x increase.

Did Claude Sonnet 5 get more expensive?

No — the opposite. The $2/$10 per million token rate was announced as introductory pricing through 31 August 2026, with an increase to $3/$15 scheduled for 1 September. Anthropic subsequently confirmed that increase will not happen and $2/$10 is now the standard price. Teams that budgeted for $3/$15 can revise downward.

How do I avoid being surprised by a price change?

Providers rarely announce increases prominently; they appear as footnotes on pricing pages, as introductory-rate expiry dates, or as billing restructures rather than headline rises. The practical defences are to re-read provider pricing pages on a schedule rather than at integration time, to record the date your cost model was built, and to treat any rate labelled introductory or promotional as temporary until proven otherwise.

Are model retirements a cost issue too?

Effectively yes. When a model retires you are forced onto its replacement, which is frequently priced differently, so a retirement date is a scheduled repricing event for anyone still on the old model. Retirements also break outright rather than degrading — a call to a retired model fails, it does not fall back.

Pricing verified 2026-10-05, retirements verified 2026-10-05, both against official provider documentation. This log records only changes a provider has published — we do not forecast.