Cohere Pricing: Plans & API Token Cost Calculator 2026

Positioning

Cohere skips the flat monthly seat and charges you per model and endpoint, which is the right structure for search-heavy work and the wrong one to eyeball. Command R7B is aggressively cheap for high-volume, low-latency calls, while the flagship Command A carries the demanding reasoning at a higher per-token rate. The lever that decides most bills is not the generate step at all, it is Rerank, priced by search unit, so a pipeline that reranks everything spends very differently from one that reranks selectively. Paired thoughtfully, R7B for volume and Command A for the hard calls, the pricing is efficient for RAG.

Cost drivers

• 1Rerank scales by search unit and multilingual rerank models can run higher per 1,000 units, so heavy reranking is where the invoice quietly grows.
• 2Fine-tuning a custom model carries a training charge per 1M tokens before you pay a cent for inference.
• 3Classify fine-tuning tasks bill separately, so a classification-heavy pipeline has its own line item.

Watch-outs

Strengths

The free trial exercises every endpoint, so you can prove the Embed and Rerank quality before paying anything.

• Command A carries enterprise RAG and tool use over your own data
• State-of-the-art multilingual embeddings across 100+ languages
• A dedicated Rerank endpoint that measurably lifts search relevance

What users say