Video output falls 30% to $12 per million tokens on the new generally available model, against $17.50 on 1.1. Input and text output prices stay the same, and paid-tier rate limits double.
Gemini API retires its embedding preview and reroutes the calls
EditorialBy TrustList Editorial
Google says the replacement keeps the same schema, pricing and embedding space, so no re-indexing is needed. Calls to the preview string now bill at Gemini Embedding 2 rates.
- United States of America
- California
- Mountain View, Ca
- Machine Learning
- +2 more
About Gemini API retires its embedding preview and reroutes the calls
Gemini API retires its embedding preview and reroutes the calls
6 November 2026: Google has deprecated gemini-embedding-2-preview, and requests to that string now go to gemini-embedding-2 at Gemini Embedding 2 rates. Code that pinned the preview keeps working, but it now runs on a different model than the name in its config, and Google asks developers to change the string.
Not yet independently verified. We read Google's changelog and deprecations table but did not test the route or read the current price of gemini-embedding-2. The claim that no re-indexing is needed is Google's. We will update this when it can be confirmed, and remove this note.
What changed in the API
The Gemini API changelog entry says the route took effect on 6 November. The deprecations table gives the preview a release date of 10 March 2026 and a shutdown date of 6 November 2026, with gemini-embedding-2 as the replacement, so the preview lasted about eight months.
The instruction is a one-line edit: update the model string to gemini-embedding-2. Google says the replacement "uses the same API schema, pricing, and embedding space", and on that basis says no other code changes or vector re-indexing are needed.
Why the embedding space claim is the one that matters
Embedding swaps are usually costly because vectors from two models cannot be compared, which forces a full re-embed of the stored corpus. If the space is the same, as Google states, vectors already in an index stay comparable with new ones from gemini-embedding-2.
A team with a large index can check that cheaply before relying on it. Embed a small sample of stored documents again through the new string and compare the results with the vectors already held. A mismatch would be worth raising with Google, because the changelog says no re-indexing is needed.
Those stored vectors tend to live in the kind of product our vector database listings cover.
What else to look for in a codebase
The table also lists older embedding models with gemini-embedding-2 as their replacement. text-embedding-004 shut down on 14 January 2026. embedding-001, embedding-gecko-001, gemini-embedding-exp and gemini-embedding-exp-03-07 shut down on 30 October 2025.
A search of configuration and environment files for any of those strings, plus the preview name, finds the calls to fix. Rerouted traffic is easy to miss because nothing fails, and the same goes for the coming Omni 1.1 Flash reroute on the same API.
Cost tracking is the other place the old name may linger. Google says calls to the preview string are charged at Gemini Embedding 2 rates, and that the two share the same pricing.
Sources
Categories & features
- United States of America
- California
- Mountain View, Ca
- Machine Learning
- Semantic Search
- Vector Database
TrustList Weekly
The week in software and IT, in one email
The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.
You may be interested in
Standalone annual subscriptions cannot renew after 31 January 2027, and new sign-ups stopped on 9 October. Google points customers to Antigravity as the replacement.
New workspaces and sign-ups stopped on 22 June. Deployed apps and core Firebase products such as Firestore and Authentication carry on, but workspaces must move to AI Studio or Antigravity.
The cash deal is due to close in 2026. Five9 will also sell Level AI standalone, including to on-premises contact centres that have no plan to buy its cloud platform.
Suplari's AI spend analytics goes into Docusign's Intelligent Agreement Management platform and its Iris engine, to flag purchases outside agreed terms. Terms were not disclosed.
Dialpad founder Craig Walker says customer contracts, pricing and integrations stay unchanged. BetaKit reports an all-share deal due to close around 12 November, with Ada's Mike Murchison as chief AI officer.
The coding-agent workspace launched in July 2025 stopped taking sign-ups on 4 November. Cloud turns read-only on 13 November, its data is deleted from 20 November, and the Mac app winds down in early 2027.
Requests naming gemini-2.5-pro will be served by gemini-3.8-flash, and gemini-2.5-flash by gemini-3.5-flash-lite, billed at the new models' rates. Both moves bring breaking API changes, and 3.8 Flash's price doubles on 1 January.
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- More United States of AmericaThe ranking for this subject
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.