Context applies to an individual query/document evaluation, not the sum of all documents in the request. This is not a generated-output token allowance.
Longer queries are truncated to this token count; query and document share the model context window.
Ranking results
Ordered document indexes and relevance scores from 0 to 1. Scores are query dependent; compare rank rather than interpreting scores as calibrated probabilities.Source for Ranking results (opens in a new tab)
Ranking controls
top_n selects the number of returned results; omitted means all results, v2 max_tokens_per_doc defaults to 4096 and truncates longer documents, v1 max_chunks_per_doc controls chunking; v1 also offers rank_fields and return_documentsSource for Ranking controls (opens in a new tab)
Version-specific controls: v1 details are documented in the separately linked v1 API. The default v2 truncation setting is not the model context length.
This fixed API response does not imply general-purpose JSON-schema constrained generation.
Request limits
For performance, v2 recommends at most 1,000 documents per request. The best-practices guide describes a 10,000 document/chunk limit for v1-style chunking.Source for Request limits (opens in a new tab)
The recommendation is not an enforced 1,000-document ceiling. Best-practices and v1 reference differ on the equality boundary and default chunk count; check the selected transport.
API key required. Use string documents; max_tokens_per_doc controls truncation. Current per-search rate not verified in the reviewed direct API pricing page.
Versioned v1 API: document strings or objects, rank_fields, return_documents and max_chunks_per_doc. The v1 reference currently documents these controls; not interchangeable with v2.
Best-practices guide says default 1 and rejects above 10,000. Equality boundary and defaults conflict across references; use explicit settings and verify transport behavior.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.
Catalog identifier published by Cohere; endpoint provisioning, limits and regional access are separate.
Regions
Not collected
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.
Caching
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
Cohere API v2
rerank-multilingual-v3.0
Not verified
No version-specific reranking cache contract verified; a shared cached_tokens metadata property does not establish support.
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
Cohere API v1
rerank-multilingual-v3.0
Not verified
No version-specific reranking cache contract verified; a shared cached_tokens metadata property does not establish support.
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
Amazon SageMaker
Not verified
No version-specific reranking cache contract verified; a shared cached_tokens metadata property does not establish support.
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
Microsoft Foundry
Cohere-rerank-v3-multilingual
Not verified
No version-specific reranking cache contract verified; a shared cached_tokens metadata property does not establish support.
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
What the measurements say
No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
The exact original-model repository publishes a tokenizer and no model weights; its README explicitly calls it a tokenizer. Cohere documents managed inference for the original model. A public tokenizer does not make the underlying model open-weight.
