BuildIndependently confirmed2 publishers2 min readPublished
Cohere says Embed 5 Fast keeps 98.4% of Pro's retrieval score when querying a Pro index
Cohere's Embed 5 lets teams index with Pro at $0.12 per million text tokens and query with Fast at $0.08 against the same vectors. The quality and throughput figures behind that split come from Cohere's own tests, run on datasets and parsing pipelines a buyer may not share.
The Engineer · Build desk
What happened
- Cohere reports that Fast processed documents at an average of 2.4 times Pro's throughput in its own tests.
- Cohere introduced RCP-nDCG@10, a scoring method in which an AI judge grades documents against query-specific relevance criteria instead of fixed labels.
- Embed 5 shipped on September 30 through the Cohere API, Model Vault, Microsoft Foundry and Amazon SageMaker, with single-tenant deployment via Model Vault.
Why it matters
- decision Since the query encoder can change without re-embedding the corpus, a team can launch on Pro end to end and move the query path to Fast later, or back, as an ordinary deploy.
- cost At Cohere's recommended 1,024-dimension int8 setting, 100 million chunks need about 102 GB, one eighth of the float32 footprint, and Fast can still query those vectors.
- constraint The rival comparisons ran on text parsed by ViDoRe's authors or by Gemini 1.5 Flash, so a team with a different parsing pipeline cannot assume the ranking holds on its own records.
Pro writes the document vectors. Fast encodes each query into the same space at the same dimensions, and the vector store compares the two directly [1][6].
On Cohere's numbers, that swap costs 1.6 points against a Pro-to-Pro baseline of 100 [21]. Running Fast on both sides costs 3.4 points. Indexing with Pro therefore buys back 1.8 of them, a little over half [21]. Cohere says no individual dataset showed a major drop under the mix [5]. It scored the cross-model test with standard nDCG@10, not with its new metric [13].
Fast's text price is a third below Pro's, or $80 per billion text tokens against $120 [19][20]. Image inputs cost $0.40 per million tokens on either model, so an image query path gets no discount [2]. A $40 gap per billion tokens will rarely be the line that gets this change through review [20].
Throughput is the stronger case. Cohere's 2.4x figure is document throughput from its own tests [7]. Cohere assigns Fast to interactive search, high-volume retrieval and agent workflows [8]. Those workloads wait on the latency of one short query. Bulk document throughput is a separate measurement. The announcement does not include a customer workload or an independent production result [15].
The rival margins are narrow. On ViDoRe V3, Pro leads Voyage 4 Large by 2.1 points and Fast leads it by 0.8 [22]. On Cohere's parsed-PDF suite, Pro scored 84.8 to 83.6 for Voyage 4 Large and 80.8 for Gemini Embedding 2 [10]. The multilingual results are mixed. Pro leads a five-language European average with 77 to Gemini Embedding 2's 73, but trails Gemini on nine of the individual tests [18].
Cohere says it validated the RCP-nDCG@10 judge against human judgment, and it also used the method to optimize the models being scored [12]. The metric grades reranking over a fixed candidate set. First-stage retrieval from the full corpus is scored separately, with standard nDCG and Recall [13]. The New Stack notes that the reported scores are not directly comparable across Cohere's evaluations [13].
For any of this to transfer, a team's queries would need to look like the 40 datasets in the cross-model test, and its documents would need to reach the embedder in a similar state [3]. The shared space makes that cheap to check. One Pro index, both encoders on the query side and the team's own labelled queries are enough to measure the 1.6-point figure on the team's own data [1][21].
What to watch
- An independent evaluation of Embed 5 against Voyage 4 Large and Gemini Embedding 2 on standard nDCG, using page images instead of parsed text.
- A production report giving per-query latency for Fast against a Pro-built index, which Cohere's document-throughput figure does not measure.
- Whether evaluators outside Cohere adopt RCP-nDCG@10, or its AI-judge scores stay confined to Cohere's own results.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+20
- Incentives75
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Embed 5 Pro and Fast share an embedding space, so customers can build an index with Pro and query it with Fast without rebuilding the index; Cohere recommends indexing with Pro and querying with Fast.
ReportedSupportedSource: Cohere, via runtimewire.com2 sources— create a free account to open themView cited source - [2]
Embed 5 Pro costs $0.12 per million text tokens and Fast costs $0.08; image inputs cost $0.40 per million tokens for either model.
ReportedSupportedSource: Cohere product announcement, via runtimewire.com2 sources— create a free account to open themView cited source - [3]
Across 40 datasets covering text, images, fused documents and parsed documents, Fast queries against a Pro index scored 98.4 relative to a Pro-to-Pro baseline of 100.
ReportedSupportedSource: Cohere testing, via The New Stack2 sources— create a free account to open themView cited source - [4]
Using Fast for both indexing and queries dropped the score to 96.6 against the Pro-to-Pro baseline of 100.
ReportedSupportedSource: Cohere testing, via The New Stack2 sources— create a free account to open themView cited source - [5]
Cohere says none of the individual datasets showed a major drop when Pro and Fast were used together.
ReportedSupportedSource: Cohere, via The New Stack2 sources— create a free account to open themView cited source - [6]
Pro and Fast produce compatible vectors at the same dimensions, and Cohere says teams can still mix the two when using Matryoshka truncation or int8 quantization.
ReportedSupportedSource: Cohere, via The New Stack2 sources— create a free account to open themView cited source - [7]
Cohere reports that Fast processed documents at an average 2.4 times Pro's throughput in its tests; the figures are company-reported.
ReportedSupportedSource: Cohere, via runtimewire.com and The New Stack2 sources— create a free account to open themView cited source - [8]
Cohere positions Pro for offline indexing and Fast for interactive search, high-volume retrieval and agent workflows.
ReportedSupportedSource: Cohere, via runtimewire.com2 sources— create a free account to open themView cited source - [9]
On ViDoRe V3, evaluated using parsed text outputs curated by the benchmark's authors rather than page images, Pro averaged 85.8 and Fast 84.5, compared with 83.7 for Voyage 4 Large, 83.2 for Gemini Embedding 2 and 77 for Cohere's previous Embed 4.
ReportedSupportedSource: Cohere-reported results, via The New Stack and runtimewire.com2 sources— create a free account to open themView cited source - [10]
On a separate parsed-document suite, Cohere reports an average of 84.8 for Pro, ahead of Voyage 4 Large at 83.6 and Gemini Embedding 2 at 80.8; the documents in that suite were parsed using Gemini 1.5 Flash. ViDoRe V3 documents were parsed as text, with evaluation material curated by ViDoRe's authors.
ReportedSupportedSource: Cohere-reported results, via runtimewire.com2 sources— create a free account to open themView cited source - [11]
Cohere introduced RCP-nDCG@10, a retrieval-scoring method used in its evaluations that grades documents against query-specific relevance criteria with an AI judge, rather than relying only on pre-existing relevance labels.
ReportedSupportedSource: runtimewire.com2 sources— create a free account to open themView cited source - [12]
Cohere says it validated RCP-nDCG@10 against human judgment; its own explanation describes the evaluation as a new approach the company developed and used to optimize its models.
ReportedSupportedSource: Cohere, via runtimewire.com2 sources— create a free account to open themView cited source - [13]
RCP-nDCG@10 measures reranking over a fixed candidate set rather than first-stage retrieval from the full corpus; first-stage retrieval is evaluated separately with standard nDCG and Recall, while the fused text-image, page-image and cross-model tests use standard nDCG@10, so the reported scores are not directly comparable across evaluations.
ReportedSupportedSource: The New Stack2 sources— create a free account to open themView cited source - [14]
Cohere released Embed 5 on September 30; it is available through the Cohere API, Model Vault, Microsoft Foundry and Amazon SageMaker, and supports single-tenant deployment through Model Vault.
ReportedSupportedSource: runtimewire.com2 sources— create a free account to open themView cited source - [15]
The announcement does not provide a customer workload or independent deployment result demonstrating the same performance in production.
ReportedSupportedSource: runtimewire.com2 sources— create a free account to open themView cited source - [16]
Cohere puts a 2,048-dimensional float32 vector at 8 KB, roughly 819 GB for 100 million chunks; a 1,024-dimensional int8 vector cuts that to about 102 GB, and a 256-dimensional binary vector to roughly 3.2 GB.
- [17]
For most deployments, Cohere recommends 1,024-dimensional int8, which it says retains close to full-precision retrieval quality.
- [18]
Pro leads Cohere's five-language European average with a score of 77, against 76 for Voyage 4 Large and 73 for Gemini Embedding 2, but trails Gemini Embedding 2 on nine of the tests.
- [19]
Fast's text-token price is one third (about 33%) below Pro's.
- [20]
Per billion text tokens, Pro costs $120 and Fast $80, a $40 gap.
- [21]
Querying a Pro index with Fast loses 1.6 points; all-Fast loses 3.4 points; indexing with Pro recovers 1.8 points, about 53% of the all-Fast loss.
- [22]
On ViDoRe V3, Pro leads Voyage 4 Large by 2.1 points and Fast leads it by 0.8 points.
- [23]
For 100 million chunks, 1,024-dimension int8 storage (about 102 GB) is one eighth of 2,048-dimension float32 storage (about 819 GB).
Sources
2 independent publishers whose own reporting we read for this story.
- runtimewire.comCohere launches Embed 5 with a cheaper model for live enterprise search
1 article · September 30, 2026
- thenewstack.ioCohere’s faster query model barely dents retrieval quality in its tests
1 article · September 30, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Entities
- RCP-nDCG@10Follow
- ViDoRe V3Follow
- Amazon SageMakerFollow
- Voyage 4 LargeFollow
- Embed 5Follow
- Embed 4Follow
- Aidan GomezFollow
- CohereFollow
- Gemini Embedding 2Follow
- Microsoft FoundryFollow