What it is
Rerank 4 is Cohere's previous-generation reranker, released on 11 December 2025 as rerank-v4.0-pro and rerank-v4.0-fast. On 21 October 2026 Cohere released Rerank 5, which its docs now list first; Rerank 4 remains on the API and no retirement date has been announced. A reranker rescores a search system's shortlist by reading the query and each document together, which is slower than vector search but more precise.
Both versions are multilingual, rank semi-structured JSON as well as plain text, and accept 32,000 tokens per query-and-document pair, up from 4,096 on Rerank 3.5. Cohere bills Rerank 4 per search: one query with up to 100 documents, where any document over 500 tokens including the query is split and each piece counted.
It is available through Cohere's API, Microsoft Foundry and gateways such as Vercel's, which lists Pro at $2.50 and Fast at $2.00 per 1,000 searches. Cohere's own pricing page now shows Model Vault rates only for Rerank 5.
Our review
Rerank 4 has been overtaken by its successor. Cohere's Rerank 5 launch post puts Rerank 5 Pro 12.4 points ahead of Rerank 4 Pro on BRIGHT, a reasoning-heavy retrieval benchmark, and says even Rerank 5 Fast beats Rerank 4 Pro on all four headline benchmarks. Those are Cohere's figures on its own RCP-nDCG@10 metric, but they come from the one vendor with a reason to flatter the older model, and they agree with Voyage's September claim that its rerank-3 led Rerank 4 Pro by 2.72% across 95 datasets.
What it did well still holds. The 32K-token window, eight times Rerank 3.5's, spares long passages from being cut into fragments, and JSON support helps with tickets, product records and table rows. Both carry over into Rerank 5, which adds calibrated scores between 0 and 1 and Context Focus, which returns the relevant spans inside each document.
The billing model is what to compare. Rerank 4 charges per search, $2.50 per 1,000 for Pro on Vercel's gateway, while Rerank 5 charges per token at $0.05 per million for Pro. A full search of 100 documents at 500 tokens is 50,000 tokens, about $0.0025 on Rerank 5 and the same as Rerank 4. A typical RAG call of 20 passages at 300 tokens comes to about $0.0003 on Rerank 5, roughly an eighth of Rerank 4's flat charge.
The case for staying is narrow: a pipeline whose cut-offs were tuned on Rerank 4 scores, and no evaluation budget yet for a switch. Rerank 5's scores are calibrated where Rerank 4's were not, so changing the model name without re-fitting thresholds will change what reaches the LLM.
Where it earns its place
A search engineer running Rerank 4 in production
Move to Rerank 5 without breaking relevance thresholds
Replay a held-out query set through both models and re-fit score cut-offs, since Rerank 5 scores are calibrated; the endpoint and request format are unchanged. At 1,000 requests a minute on a production key, a 50,000-query replay takes under an hour.
A finance lead budgeting a support-search assistant at 2 million queries a month
Forecast reranking spend
On Rerank 4 Pro at $2.50 per 1,000 searches that is $5,000 a month however short the documents. On Rerank 5 Pro, 20 passages of 300 tokens per query is 12 billion tokens, about $600.
An Azure-standardised enterprise IT group
Add reranking inside Microsoft Foundry under existing contracts
Rerank models run in the Foundry catalogue, and Cohere says it receives no prompts on third-party cloud platforms, so data stays under the Azure agreement. New projects there have little reason to start on version 4 now that Rerank 5 is listed too.
Before you commit
- Setup
- Same Rerank endpoint as Rerank 5: pass rerank-v4.0-pro or rerank-v4.0-fast with a query and documents.
- Integrations
- Cohere API, Microsoft Foundry and gateways such as Vercel AI Gateway. Model Vault pricing is now published only for Rerank 5.
- Data and privacy
- On Cohere's SaaS, logs are deleted after 30 days and training use is opt-out under Data Controls; on cloud platforms and private deployments Cohere receives no inputs. SOC 2 Type II and ISO 27001.
- Cost at scale
- $2.50 per 1,000 searches for Pro and $2.00 for Fast via Vercel, so 2 million searches a month is $5,000 on Pro. Documents over 500 tokens count as several, so 100 long documents can exceed one search unit.
- Limits
- 32K tokens per query-document pair; text only. Production keys allow 1,000 rerank requests a minute, trial keys 10. Rerank fine-tuning was retired in September 2025. No retirement date published for Rerank 4.
What's good
- 32K-token context per query-document pair
- Ranks semi-structured JSON
- Flat per-search billing is easy to forecast
- On the Cohere API, Microsoft Foundry and gateways
What's not
- Superseded by Rerank 5 on 21 October 2026
- Behind Rerank 5 Fast and Voyage rerank-3 in vendor tests
- Per-search price no longer on Cohere's pricing page
- Documents over 500 tokens are split and billed as several
Best for
Teams with a tuned Rerank 4 pipeline who need time before re-validating on Rerank 5.
Alternatives
Cohere Rerank 4: common questions
- Is Cohere Rerank 4 worth it?
- Rerank 4 still works, but Cohere's own Rerank 5 now beats it on every benchmark Cohere reports, which makes it a model to migrate from rather than adopt. We rate it 3.7 out of 5.
- How much does Cohere Rerank 4 cost?
- Billed per search (one query, up to 100 documents). Vercel AI Gateway lists Pro at $2.50 and Fast at $2.00 per 1,000 searches; Cohere's page no longer shows Rerank 4 rates.
- What are the downsides of Cohere Rerank 4?
- Superseded by Rerank 5 on 21 October 2026; behind Rerank 5 Fast and Voyage rerank-3 in vendor tests; per-search price no longer on Cohere's pricing page; documents over 500 tokens are split and billed as several.
- Who is Cohere Rerank 4 best for?
- Teams with a tuned Rerank 4 pipeline who need time before re-validating on Rerank 5.
- How much does Cohere Rerank 4 cost for a team or in production?
- $2.50 per 1,000 searches for Pro and $2.00 for Fast via Vercel, so 2 million searches a month is $5,000 on Pro. Documents over 500 tokens count as several, so 100 long documents can exceed one search unit.
- Is Cohere Rerank 4 safe for company data?
- On Cohere's SaaS, logs are deleted after 30 days and training use is opt-out under Data Controls; on cloud platforms and private deployments Cohere receives no inputs. SOC 2 Type II and ISO 27001.
- What are the best Cohere Rerank 4 alternatives?
- The closest alternatives we have reviewed are Cohere Rerank 5, Voyage rerank-3, Jina Reranker v3.5 and Qwen3-Embedding and Qwen3-Reranker.
Answers checked Oct 23, 2026.