Cohere Rerank 4
Cohere's 32K-context reranker in Pro and Fast flavors — built so AI agents can rank whole documents, not fragments.
About
Rerank 4 is Cohere's 2025 refresh of its flagship reranker, and the headline change is context: 32K tokens (32,768), quadruple Rerank 3.5's 4K. Documents are chunked at up to ~32,764 tokens — roughly 48 to 50 pages — so the model ranks entire documents rather than fragments, which is aimed squarely at agentic search where the agent needs confidence in long, messy enterprise documents.
It ships in two tiers on the v2 /rerank endpoint: rerank-v4.0-pro for deep reasoning and precision work (risk models, analysis), and rerank-v4.0-fast, a smaller model for latency-sensitive jobs like e-commerce and customer service. Both cover 100+ languages, use a cross-encoder that scores query and candidates jointly, and are available on Cohere's API, AWS Bedrock/SageMaker, and Microsoft Foundry.
It is brand new (launched December 2025), so independent BEIR-style evaluations have not been published yet; the quality story is entirely Cohere's own testing so far.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No independent published benchmarks as of this entry — the model launched December 2025 and third-party BEIR/MTEB-rerank evaluations have not appeared yet. Cohere-reported (blog + VentureBeat coverage): 4x context window over 3.5 (4K → 32K) with improved accuracy for agentic enterprise search; internal enterprise figures only.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: Cohere launch (cohere.com/blog/rerank-4, Dec 2025) with VentureBeat coverage and Microsoft Foundry listing (public preview Dec 11 2025); pricing ($2.50/1K Pro, $2.00/1K Fast) corroborated across Microsoft docs and third-party pricing trackers. Strengths: major vendor, concrete spec disclosures (32,768 context, chunking behavior documented in Fern-hosted Cohere docs), real deployment footprint. Docked for: (1) no independent benchmarks published yet — quality claims are vendor-only; (2) closed weights, API-only. License: proprietary.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Launch 2025-12-11 verified via cohere.com/blog/rerank-4, VentureBeat coverage, and Microsoft Foundry listing (public preview Dec 11 2025).
- · Specs (rerank-v4.0-pro / rerank-v4.0-fast, 32,768 context, 100+ languages, cross-encoder) corroborated across three sources; pricing $2.50/1K (Pro) and $2.00/1K (Fast) per search unit.
- · No independent benchmarks published yet — quality claims are vendor-only. Closed weights, API-only.