Cohere model deprecations — retirement dates and replacements

No Cohere models currently have an announced retirement date. 14 have already been retired.

Which Cohere models are being retired, and when?

No retirements are currently scheduled.

How much notice does Cohere give before shutting a model down?

Across 14 Cohere models where both the announcement and the shutdown date are published, the median gap is 0 days. The range runs from 0 to 149 days, and the tightest case on record is embed-english-v2.0 at 0 days (2026-04-04 → 2026-04-04). That number is not published by Cohere or anyone else — it is computed here from their own dates, and it is what tells you how much runway a pinned model version really has. Same figure for every provider →

How fast is the Cohere API once you migrate?

Knowing the shutdown date only solves half the problem. These are our own measurements of the live Cohere API, so you can see what the endpoint you are migrating to actually costs you in latency, by region:

Regionp50 TTFBUptime
Asia (Tokyo)174 ms100%
Europe (Germany)200 ms100%
South America (São Paulo)159 ms100%
US (Central)78 ms100%

Edge latency over the last 24 hours. Full Cohere latency detail →

Which Cohere models have already been retired?

ModelRetiredReplacementSource
embed-english-v2.02026-04-04embed-english-v3.0source
embed-english-light-v2.02026-04-04embed-english-v3.0source
embed-multilingual-v2.02026-04-04embed-multilingual-v3.0source
c4ai-aya-expanse-8b2026-04-04command-r7b-12-2024source
c4ai-aya-vision-8b2026-04-04command-r7b-12-2024source
command-r-03-20242025-09-15command-r-08-2024source
command-r-plus-04-20242025-09-15command-r-plus-08-2024source
command-light2025-09-15command-r-08-2024source
command2025-09-15command-r-08-2024source
summarize2025-09-15source
command-r2025-09-15command-r-08-2024source
command-r-plus2025-09-15command-r-plus-08-2024source
rerank-english-v2.02025-04-30rerank-v3.5source
rerank-multilingual-v2.02025-04-30rerank-v3.5source

How is this calendar maintained?

A scheduled job re-reads Cohere’s official deprecation page every week and rewrites this table from it. Entries that cannot be traced to a published source are not shown at all, which is why every row carries a source link. Dates change: providers extend deadlines and occasionally bring them forward, so this page reflects what the vendor documented at the last check, not a prediction.

All providers: full deprecation calendar · Latency data under CC BY 4.0: JSON API