Cohere Rerank v3.5 · inference-research

Fields

id

models/cohere-rerank-v3-5

Family

cohere-rerank

Vendor

Cohere

Release year

2024

Architecture

dense

Max context

4,096

Attention variant

cross-encoder (architecture not specified)

KV cache (bytes/token)

0

Open weights

no

License

Proprietary

notes

Cohere ships rerank-v3.5 as a closed API; the HF repo Cohere/Cohere-Rerank-v3.5 contains no config.json with transformer geometry. The KV-geometry fields are left null rather than guessed. RERANKER: cross-encoder. Context length 4096 tokens (query+document combined). Supports same languages as embed-multilingual-v3.0. Not open weights — available via Cohere API and AWS Bedrock. No config.json available on HF (only tokenizer). Architecture details not publicly specified. The 4096-token context limit is the serving-relevant fact. CONTEXT_SCALING_METHOD (null): proprietary API; record's own attention_variant is 'architecture not specified', so no positional scheme is established either way.

task

reranking

Notes

Cohere ships rerank-v3.5 as a closed API; the HF repo Cohere/Cohere-Rerank-v3.5 contains no config.json with transformer geometry. The KV-geometry fields are left null rather than guessed. RERANKER: cross-encoder. Context length 4096 tokens (query+document combined). Supports same languages as embed-multilingual-v3.0. Not open weights — available via Cohere API and AWS Bedrock. No config.json available on HF (only tokenizer). Architecture details not publicly specified. The 4096-token context limit is the serving-relevant fact.

CONTEXT_SCALING_METHOD (null): proprietary API; record's own attention_variant is 'architecture not specified', so no positional scheme is established either way.

Cross-references

This record references no other records — it is a leaf. It is cited by 0.