Cohere
AI platform providing large language models and NLP tools via API for enterprises.
Cohere is shipping a model a month and retiring the old ones just as briskly.
◆Recent moves
- 1mo ago
Meet Cohere Transcribe Arabic
Cohere Transcribe gains an Arabic model, extending the speech line launched earlier in the year. It broadens language coverage rather than changing what the transcription product does. The entry carries a single line of detail, so accuracy or deployment specifics are not visible.
View source ↗ - 2mo ago
Announcing Cohere’s North-Mini-Code-1.0
A code-focused model, North-Mini-Code-1.0, joins the lineup alongside the Command and Rerank families. The naming places it under Cohere's North product rather than the general Command line, which fits the pattern of purpose-built models sitting beneath the flagship. The entry was captured as a headline with no body, so capabilities and sizing are unknown from this feed.
View source ↗ - 2mo ago
Announcing Cohere’s Command A+
Command A+ arrives nine days after Command A Reasoning, continuing a rapid cadence on the flagship line. On naming alone it reads as a step up from the Command A generation. No body text was captured, so the actual changes in capability, context length or pricing cannot be assessed from this entry.
View source ↗ - 3mo ago
Cohere's Rerank v4.0 Model is Here!
⚡ SPARKRerank 4.0 ships in two variants — a pro model tuned for ranking quality and a fast model tuned for latency and throughput. Splitting a single model into a quality-versus-speed choice is a change in how Cohere packages capability, not just an accuracy bump. Reranking is the piece of the retrieval stack Cohere sells most directly into other companies' pipelines.
- 3mo ago
Announcing Cohere's Command A Reasoning Model
⚡ SPARKCommand A Reasoning is a 111-billion-parameter hybrid reasoning model with a 256K context window, aimed at complex agentic tasks across English and 22 other languages. It moves Cohere's flagship line from general instruction-following toward deliberate reasoning. The multilingual scope is the part that distinguishes it from most reasoning models arriving this year.
- 4mo ago
Retirement of Embed v2.0 and Aya Expanse / Vision 8B
Five models — the Embed v2.0 family and the Aya Expanse and Vision 8B models — stop serving entirely, with requests to those IDs failing from 4 April. This is a hard cutoff rather than a deprecation warning, so it carries real migration work for anyone still pinned to them. It is the maintenance cost of the release cadence visible elsewhere in this changelog.