What it is
Embed 5 is Cohere's newest embedding family, launched on 30 September 2026 as embed-v5.0-pro, for retrieval quality, and embed-v5.0-fast, for latency and cost. Embedding models turn content into vectors so software can find related material, the basis of search and retrieval-augmented generation (RAG).
Both take text, images or fused text-and-image input such as PDF pages, accept up to 128K tokens per input and were trained on more than 100 languages. Output runs from 256 to 2,048 dimensions, as float, int8 or binary. Pro and Fast share one embedding space, so an index built with Pro can be queried with Fast.
It is sold on Cohere's API at $0.12 per million text tokens for Pro and $0.08 for Fast, with images at $0.40 per million on both, and through Microsoft Foundry, Amazon SageMaker and Cohere's dedicated Model Vault from $3 an hour.
Our review
Embed 5 belongs on the shortlist for long, visually dense business documents. A 128K-token input lets whole filings and reports go in as one piece, and page-image and fused text-image input mean slides and scanned tables need not be flattened to text first. Cohere's launch post puts Embed 5 Pro at 85.8 on ViDoRe V3, ahead of Voyage 4 Large (83.7), Gemini Embedding 2 (83.2) and OpenAI's text-embedding-3-large (75.5), and 8.8 points above Embed 4.
Read those numbers with care. They use RCP-nDCG@10, a metric Cohere devised to score results against query-specific relevance criteria rather than fixed labels, and the margins over Voyage are one to two points. The widest gap is on fused text-image corpora, where Cohere reports 82.3 for Pro against 61.3 for Gemini Embedding 2. We found no independent leaderboard result for Embed 5 yet.
Pricing, missing from the docs at first, is now public and competitive. Indexing 5 billion text tokens, roughly ten million pages at 500 tokens each, costs about $600 with Pro or $400 with Fast. Images cost $0.40 per million tokens on either tier, more than three times the Pro text rate. The shared space between tiers is the practical gain: index once with Pro, then serve queries with the cheaper Fast without re-embedding.
Rate limits are the catch for bulk jobs. A production key allows 2,000 text inputs a minute but only 400 images, and the docs list just the synchronous Embed endpoint for v5, not the batch Embed Jobs. A million page images is about 42 hours of continuous calls at that ceiling, which is roughly where a $2,000-a-month Model Vault instance starts to earn its keep.
Where it earns its place
A search engineer at an asset manager indexing 20 years of annual reports
Retrieve the right table from 300-page filings
Embed page images or fused page-and-metadata vectors with Pro, then query the same index with Fast. Cohere reports Pro first on FinanceBench (80.1) and FinQA (90.0) in its own tests. At $0.12 per million text tokens, a 5-billion-token archive costs about $600 to index.
A head of knowledge management at a 2,000-person engineering firm
Make scanned manuals and slide decks searchable without a parsing project
Page-image embedding skips OCR clean-up for diagrams and tables. For data control the same model runs on Microsoft Foundry, SageMaker or a private deployment, where Cohere says it receives no inputs; on Cohere's SaaS, switch training off under Data Controls.
A RAG developer at a 15-person startup
Keep interactive search fast and storage small
Use embed-v5.0-fast at $0.08 per million tokens; int8 output cuts vector storage four-fold against float and binary 32-fold. Trial keys are free but capped at 1,000 calls a month and barred from production use.
Before you commit
- Setup
- Sign up for a free trial key, then apply for a production key from the billing page; call the Embed endpoint with embed-v5.0-pro or embed-v5.0-fast.
- Integrations
- Cohere Embed API, Microsoft Foundry, Amazon SageMaker, Model Vault, and private VPC or on-premises deployment.
- Data and privacy
- On Cohere's SaaS platform, logged prompts are deleted after 30 days and training use can be switched off under Data Controls; on cloud platforms and private deployments Cohere receives no inputs. SOC 2 Type II, ISO 27001 and ISO 42001 per Cohere's trust centre.
- Cost at scale
- 5 billion text tokens cost about $600 on Pro or $400 on Fast; image tokens $0.40 per million. Model Vault instances run $3/hour or $2,000/month (small) and $5/hour or $3,250/month (medium).
- Limits
- 128K tokens per input. Production keys: 2,000 text inputs or 400 image inputs a minute; trial keys 5 image inputs a minute and 1,000 calls a month. Batch Embed Jobs are listed only for v3 models.
What's good
- 128K-token input per item
- Page-image and fused text-image embedding
- Pro and Fast share one vector space
- Published prices: $0.12 and $0.08 per 1M text tokens
What's not
- Benchmarks are Cohere's own, on its own metric
- Images limited to 400 inputs a minute on production keys
- Image input costs $0.40 per 1M tokens on both tiers
- Training use on the SaaS platform is opt-out
Best for
Enterprises indexing long PDFs, filings and slide decks with mixed text and images.
Alternatives
Cohere Embed 5: common questions
- Is Cohere Embed 5 worth it?
- Built for long, visually messy business documents and now priced in the open, though every benchmark that flatters it is Cohere's own. We rate it 4.2 out of 5.
- How much does Cohere Embed 5 cost?
- API: Pro $0.12 and Fast $0.08 per 1M text tokens; images $0.40 per 1M. Model Vault from $3/hour or $2,000/month per instance.
- What are the downsides of Cohere Embed 5?
- Benchmarks are Cohere's own, on its own metric; images limited to 400 inputs a minute on production keys; image input costs $0.40 per 1M tokens on both tiers; training use on the SaaS platform is opt-out.
- Who is Cohere Embed 5 best for?
- Enterprises indexing long PDFs, filings and slide decks with mixed text and images.
- How much does Cohere Embed 5 cost for a team or in production?
- 5 billion text tokens cost about $600 on Pro or $400 on Fast; image tokens $0.40 per million. Model Vault instances run $3/hour or $2,000/month (small) and $5/hour or $3,250/month (medium).
- Is Cohere Embed 5 safe for company data?
- On Cohere's SaaS platform, logged prompts are deleted after 30 days and training use can be switched off under Data Controls; on cloud platforms and private deployments Cohere receives no inputs. SOC 2 Type II, ISO 27001 and ISO 42001 per Cohere's trust centre.
- What are the best Cohere Embed 5 alternatives?
- The closest alternatives we have reviewed are voyage-4-large, Gemini Embedding 2 and text-embedding-3-large.
Answers checked Oct 22, 2026.
Make Cohere Embed 5? Show this rating on your site
