Cohere

Current flagship; 256K context, optimized for multi-step agentic workflows, complex RAG with tool use, and structured outputs. $2.50/$10.00 per 1M.

Open-weights 7B model (Apache 2.0); cheapest hosted option at $0.04/$0.15 per 1M; most accessible Cohere model for self-hosting.

Cost-efficient default for production RAG; 128K context. $0.15/$0.60 per 1M (70% cut from launch pricing).

Proven production RAG workhorse with grounded generation and citations; 128K context. Superseded by Command A for complex/agentic tasks but still active. $2.50/$10.00 per 1M.

Original flagship RAG model; better reasoning/instruction-following than R, 100+ language support. Launched $3.00/$15.00; superseded by the Aug 2024 update.

First Cohere model explicitly designed for RAG; introduced grounded generation with structured citations. Launched $0.50/$1.50; superseded by the Aug 2024 update.

Open-weight (Apache 2.0) multilingual model covering 23 languages; from the Aya initiative. $0.50/$1.50 per 1M. Released in 2024, exact month not stated.

Embedding model for semantic search/clustering; Matryoshka support (1024/512/256 dims). $0.10 per 1M tokens. Released year known, exact month not stated.

Multilingual embedding model covering 100+ languages for retrieval. $0.10 per 1M tokens. Released year known, exact month not stated.

Cross-encoder reranker that re-scores retrieved documents to improve RAG precision; model-agnostic. $2.00 per 1K queries. Released year known, exact month not stated.