Cohere Rerank Pricing Calculator
Estimate Cohere Rerank pricing for Rerank 4 Fast or Rerank 4 Pro from monthly queries, documents per query, document length, and a growth buffer. The calculator converts long-document chunking into billable search units instead of incorrectly treating Rerank as token-priced generation.
Quick answer · Official Cohere pricing checked 2026-07-24
Cohere Rerank pricing is $2.00 per 1,000 searches for Rerank 4 Fast and $2.50 per 1,000 searches for Rerank 4 Pro.
One search unit covers one query with up to 100 documents. If a document plus the query exceeds 500 tokens, Cohere splits it into chunks; those chunks count as separate documents and can push a request into multiple search units.
Cohere Rerank cost calculator
Enter an average workload. The estimate uses Cohere's published search-unit rule and rounds each query up to a whole number of search units.
How to use the Cohere Rerank pricing calculator
- Select Fast or Pro. Fast costs less and prioritizes latency; Pro costs more and prioritizes ranking quality.
- Enter monthly query volume. Count every production request that calls the Rerank endpoint, including retries and parallel retrieval branches.
- Enter documents per query. Use the candidate set sent to Rerank, not the smaller number returned to the LLM.
- Estimate document and query length. Length affects how many 500-token chunks Cohere counts as documents.
- Add a buffer and surrounding costs. Embeddings, vector search, generation, observability, cloud infrastructure, and taxes are separate from the Rerank API estimate.
Cohere Rerank 4 pricing comparison
These first-party API rates were verified from Cohere's pricing page on 2026-07-24. Model Vault dedicated deployment uses separate instance pricing and is not calculated here.
| Model | API price | Cost per search | Planning fit |
|---|---|---|---|
| Rerank 4 Fast | $2.00 / 1K searches | $0.0020 | Lower-cost option optimized for high performance and the lowest latency. |
| Rerank 4 Pro | $2.50 / 1K searches | $0.0025 | Higher-accuracy option with state-of-the-art performance and low latency. |
Cohere Rerank cost examples
50 documents, short chunks
100,000 queries with 50 documents of 350 tokens plus a 20-token query stay within one search unit per query.
Before growth buffer and other RAG services.
100 long documents
100,000 queries with 100 documents of 700 tokens plus the query create two chunks per document, so each request uses two search units.
Long documents can double cost even when the original document count stays at 100.
101 short documents
Only one document above the 100-document boundary makes each query consume two whole search units.
Reducing the candidate set to 100 can materially change the invoice.
What changes a Cohere reranker bill?
Candidate-set size
Billing steps up at each 100 effective documents. Retrieve broadly only when the quality gain justifies another search unit.
Chunking
Documents over the 500-token rule are split. Chunk before calling Rerank and inspect the effective list length rather than counting source files.
Retries and fan-out
Retries, query rewrites, multi-index search, and agent branches can call Rerank more than once per user request.
How to forecast and control Cohere Rerank cost
A planning estimate becomes more reliable when it starts with production traces instead of a single guessed request. Sample a representative week, record original documents, post-chunk documents, Rerank calls per user action, retries, selected model, and the API response's billed_units.search_units. Use the median for a normal scenario and a high percentile for a budget scenario.
| Control | What to measure | Cost effect | Quality check |
|---|---|---|---|
| Candidate cap | Effective documents before Rerank | Staying at or below 100, 200, or another billing boundary avoids rounding into the next search unit. | Compare recall and final-answer accuracy before lowering the cap. |
| Chunk size | Chunks created per source document | Better chunking can reduce counted documents, but oversized chunks may still cross the 500-token rule. | Track retrieval relevance, not chunk count alone. |
| Call fan-out | Rerank calls per user request | Query rewrites, multiple indexes, agent branches, and retries multiply searches even when traffic is flat. | Deduplicate candidates and cache only when freshness requirements allow it. |
| Model routing | Fast versus Pro request share | Fast saves $0.50 per 1,000 searches compared with Pro at published API rates. | Route harder queries to Pro only after an evaluation shows a meaningful ranking gain. |
Reconcile the estimate monthly by dividing the Cohere Rerank charge by the published model rate to infer approximate billed searches, then compare that figure with logged search units. A gap can reveal retries, untracked services, unexpected chunk expansion, traffic growth, a different deployment channel, or a price change. Keep the rate and verification date beside each forecast so an old spreadsheet is not mistaken for a current quote.
Calculator limitations
- The chunk estimate uses one average document length. Real documents may produce a different billable distribution.
- The formula is a planning approximation of Cohere's published definition; actual billed units in the API response and invoice are authoritative.
- Trial API keys are free but rate-limited and cannot be used for production or commercial workloads.
- Model Vault, cloud-marketplace deployments, negotiated enterprise terms, taxes, currency conversion, and private deployment are excluded.
- Embedding, retrieval, vector database, LLM generation, networking, storage, logging, and application infrastructure are separate costs.
Official Cohere pricing sources
- Cohere pricing — current Fast and Pro API rates, search-unit definition, trial-key rules, and Model Vault pricing.
- How Cohere pricing works — explains that Rerank is billed by searches rather than generated tokens.
- Cohere Rerank documentation — endpoint behavior and implementation guidance.
For broader provider costs, review the Cohere pricing overview, compare models in the LLM cost calculator, or estimate hosted retrieval with the AWS Bedrock pricing calculator.
Cohere Rerank pricing FAQ
How much does Cohere Rerank cost?
Cohere lists Rerank 4 Fast at $2.00 per 1,000 searches and Rerank 4 Pro at $2.50 per 1,000 searches for the first-party API.
What counts as one Cohere Rerank search unit?
One search unit is one query with up to 100 documents after Cohere applies its long-document chunking rule.
Does Cohere Rerank pricing depend on tokens?
Rerank is priced by searches, not input and output tokens. Tokens still matter indirectly because a document plus the query that exceeds 500 tokens is split into multiple counted documents.
Is Cohere Rerank free?
A Cohere trial API key is free but rate-limited and is not permitted for production or commercial use. Production keys are pay as you go.
Should I choose Rerank 4 Fast or Rerank 4 Pro?
Choose Fast when latency and lower unit cost dominate. Choose Pro when the value of higher ranking quality justifies the extra $0.50 per 1,000 searches; validate both on your own retrieval evaluation set.
Why is my Rerank cost higher than queries multiplied by the listed rate?
Requests with more than 100 effective documents consume multiple search units. Long documents, query rewrites, retries, and multiple Rerank calls per user request can also increase billed searches.