---
title: "Cohere Pricing 2026: Command, Embed and Rerank Costs"
url: "https://dupple.com/learn/cohere-pricing"
description: "Cohere pricing in 2026: Command R at $0.15/$0.60 per 1M tokens, Embed 5 from $0.08, Rerank 5 from $0.02, Model Vault from $3/hour, and what is now sales-only."
published: "2026-11-16"
updated: "2026-11-16"
llms_txt: "https://dupple.com/llms.txt"
---
# Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only

Trusted by

810,000+

Techpresso subscribers

·

Editorial policy

·

How we make money

Written by

Louis Corneloup

Founder at Dupple, covering AI tools and strategies for 810K+ readers.

How our guides are researched

.

November 16, 2026

· Updated November 2026

10 min read

Short answer: the Cohere API is pay-as-you-go per million tokens. **Command R costs $0.15 per 1M input tokens and $0.60 per 1M output**, Command R7B $0.0375 and $0.15, Embed 5 starts at $0.08 per 1M tokens and Rerank 5 at $0.02. Dedicated Model Vault instances start at $3.00 an hour. Checked on Cohere's pricing page, 30 October 2026.

The catch is what is missing. Cohere's newest chat model, Command A+, has no paid per-token rate on the page: it is free to try and free to download, and production API use goes through sales. North and Compass are quote-only too. So the cheap, published part of Cohere is retrieval (embed, rerank, parse) and the older Command R line, while the flagship sits behind a sales call.

## Cohere pricing at a glance

*Cohere pricing at a glance: Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only. Last updated November 2026.*

| Model or product | What it does | Price | How it bills |
| --- | --- | --- | --- |
| Command A+ | Flagship chat and agent model, open weights | Free trial key and free download; production via sales | Contact sales |
| Command R | Midsize chat model, 128K context | $0.15 input, $0.60 output | Per 1M tokens |
| Command R7B | Smallest chat model, 128K context | $0.0375 input, $0.15 output | Per 1M tokens |
| Embed 5 Fast | Lightweight embeddings | $0.08 | Per 1M tokens |
| Embed 5 Pro | Top embeddings, text and image | $0.12 | Per 1M tokens |
| Rerank 5 Fast | Low-latency reranking | $0.02 | Per 1M tokens |
| Rerank 5 Pro | Highest-accuracy reranking | $0.05 | Per 1M tokens |
| Parse 5 | Document parsing and OCR | $1.50 | Per 1K pages |
| Transcribe | Speech to text | From $3.75 an hour | Per instance, via Model Vault |
| Model Vault | Dedicated model instances | $3.00 to $7.00 an hour | Per instance, hourly or monthly |
| North, Compass | Agent platform and enterprise search | Custom enterprise pricing | Contact sales |

Read on [Cohere's pricing page](https://cohere.com/pricing) on 30 October 2026, across the Workplace systems, Generative models and Advanced retrieval models tabs. Prices in US dollars.

**Dupple data:** of the 22 paid rates on Cohere's pricing page on 30 October 2026, only 7 are current pay-as-you-go API models; 9 are per-instance hourly prices and 6 are legacy or research rates. That is our count from the page, and it explains why third-party price lists for Cohere disagree so often.

## How much does Cohere's Command cost?

Command pricing now splits into three groups, and only one of them is a normal public price list.

The published group is Command R and Command R7B. **R7B is the cheapest chat model Cohere lists, at $0.0375 per 1M input tokens**, which makes it a fit for classification, routing and short answers where quality matters less than volume. Command R costs four times as much on both input and output and is the safer default for RAG answers that a customer will read.

The second group is Command A+. Its card on the pricing page reads "Free", with a $0 API key and a $0 model download. In practice that means two things: you can call it on a free trial key, and you can download the Apache 2.0 weights and run them yourself (Cohere says it runs on two H100 GPUs). What you cannot do is buy production API calls at a listed rate. Cohere's [rate-limit docs](https://docs.cohere.com/docs/rate-limits) show Command A+, Command A Reasoning, Command A Translate and Command A Vision with "Contact sales" in the production column, and note that production keys work like trial keys on these newer variants.

The third group is legacy pricing for existing customers, listed in the pricing page FAQ:

*How much does Cohere's Command cost?: Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only. Last updated November 2026.*

| Legacy model | Input per 1M tokens | Output per 1M tokens |
| --- | --- | --- |
| Command | $1.00 | $2.00 |
| Command-light | $0.30 | $0.60 |
| Command R 03-2024 | $0.50 | $1.50 |
| Command R+ 04-2024 | $3.00 | $15.00 |
| Command R+ 08-2024 | $2.50 | $10.00 |
| Aya Expanse 8B and 32B (research) | $0.50 | $1.50 |

Read on [Cohere's pricing page](https://cohere.com/pricing), 30 October 2026.

Many 2026 articles still quote that last Command R+ row as if it were Cohere's current flagship price, but Cohere labels it a legacy rate "for existing customers", and its [deprecations page](https://docs.cohere.com/docs/deprecations) marked `command`, `command-light`, `command-r-03-2024` and `command-r-plus-04-2024` as deprecated in September 2025. If you are starting a new project, budget on Command R, R7B or a sales quote for Command A+, not on these rows.

## What happened to Command A's public price?

Command A was the flagship with a public price until spring 2026. An archived copy of the pricing page from [14 May 2026](https://web.archive.org/web/20260514063043/https://cohere.com/pricing) still lists Command A at $2.50 per 1M input tokens and $10.00 per 1M output tokens. The copy from [14 June 2026](https://web.archive.org/web/20260614134301/https://cohere.com/pricing) replaces that card with Command A+ marked "Free".

So the move was not a price cut or a price rise: **the flagship went from a listed rate to an open-weights release plus a sales conversation**. For a small team, that is a real change. You can no longer estimate a Command A-class bill from the website. Your options are to self-host the free weights, ask sales for a production quote, or use Command R and accept a smaller model.

## How much do Embed, Rerank and Parse cost?

Retrieval is where Cohere still publishes clear, low per-token prices, and it is the part most teams buy.

**Embed 5 Fast costs $0.08 per 1M tokens and Embed 5 Pro $0.12.** Fast shares Pro's context window and embedding space, so you can embed a large archive on Fast and still compare vectors with Pro later. Pro adds the strongest retrieval quality and handles text, image and fused inputs across 100+ languages.

Rerank 5 now bills per token as well: $0.02 per 1M tokens on Fast and $0.05 on Pro. Rerank 5 Pro has a 32,768-token context window and a feature Cohere calls Context Focus for passage-level relevance. Parse 5, the document parser, costs $1.50 per 1,000 pages and handles tables, forms and handwriting.

The rerank switch matters if you budgeted on the old model. Archived copies up to [7 October 2026](https://web.archive.org/web/20261007073648/https://cohere.com/pricing) priced Rerank 4 Fast at $2.00 and Rerank 4 Pro at $2.50 per 1,000 searches, where one search is a query with up to 100 documents. Here is how the new token price compares (our arithmetic):

*How much do Embed, Rerank and Parse cost?: Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only. Last updated November 2026.*

| Search size | Tokens per search | Rerank 5 Fast per 1K searches | Rerank 5 Pro per 1K searches | Old Rerank 4 Pro per 1K searches |
| --- | --- | --- | --- | --- |
| 100 docs of 500 tokens | about 50,000 | $1.00 | $2.50 | $2.50 |
| 20 docs of 300 tokens | about 6,000 | $0.12 | $0.30 | $2.50 |

Read the pattern: **a full 100-document search costs the same on Rerank 5 Pro as it did on Rerank 4 Pro, but a small search is now far cheaper**, because you no longer pay for document slots you do not fill. Most RAG pipelines rerank 20 to 50 chunks, so most will see their rerank bill drop.

## What does a real Cohere bill look like?

Take a support assistant that answers questions over a help center. Each month it embeds 50M tokens of new and changed docs, reranks 5,000,000 small searches of about 6,000 tokens each, and generates answers from 1B input tokens and 200M output tokens. All totals below are our arithmetic from the list prices above.

*What does a real Cohere bill look like?: Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only. Last updated November 2026.*

| Line item | Volume | Option A | Option B |
| --- | --- | --- | --- |
| Embeddings | 50M tokens | Embed 5 Fast: $4.00 | Embed 5 Pro: $6.00 |
| Reranking | 30B tokens (5M searches) | Rerank 5 Fast: $600 | Rerank 5 Pro: $1,500 |
| Generation | 1B in, 200M out | Command R7B: $67.50 | Command R: $270 |
| **Monthly total** |  | **$671.50** | **$1,776** |

Two lessons follow for anyone budgeting a RAG app. First, **embedding is close to free at this scale**; the money goes to reranking and generation. Second, the gap between R7B and Command R is only about $200 a month here, so test answer quality on Command R before you settle for the smaller model to save money.

For comparison, the same 1B input and 200M output tokens on the legacy Command R+ 08-2024 rate would cost $4,500 a month (our arithmetic from the legacy table above). That is why pages still quoting R+ make Cohere look far pricier than its current published models are.

## How does Model Vault pricing work?

Model Vault is Cohere's dedicated, fully managed hosting: your own instance of a model, no shared resources, priced per instance per hour or per month. Cohere says billing can be hourly or through monthly or annual commitments, with "Fixed or Flex" plans; the annual rate is not on the page.

*How does Model Vault pricing work?: Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only. Last updated November 2026.*

| Model | Performance tier | Hourly rate per instance | Monthly rate per instance |
| --- | --- | --- | --- |
| Embed 5 Fast | Small | $3.00 | $2,000 |
| Embed 5 Fast | Medium | $5.00 | $3,250 |
| Embed 5 Pro | Small | $3.00 | $2,000 |
| Embed 5 Pro | Medium | $5.00 | $3,250 |
| Rerank 5 Fast | Medium | $4.00 | $2,500 |
| Rerank 5 Pro | Medium | $4.00 | $2,500 |
| Parse 5 | Medium | $4.00 | $2,500 |
| Parse 5 | XL | $7.00 | $4,300 |

Read on [Cohere's pricing page](https://cohere.com/pricing), 30 October 2026.

The monthly rate is the better deal for an always-on instance. A Small instance at $3.00 an hour for a 730-hour month comes to $2,190, against $2,000 on the monthly rate; Parse 5 XL comes to $5,110 hourly against $4,300 monthly (our arithmetic, assuming 730 hours). Hourly billing makes sense for bursts, such as a weekend backfill.

When does a dedicated instance beat pay-as-you-go? Divide the monthly rate by the per-token price. **An Embed 5 Pro Small instance at $2,000 a month breaks even at about 16.7B tokens a month**; Embed 5 Fast needs 25B, and Rerank 5 Pro needs 50B tokens to justify $2,500 (our arithmetic). Few teams reach those volumes. The real reasons to buy Vault are isolation, guaranteed throughput and compliance, not price.

Some Vault prices fell this autumn. An archived copy from [2 August 2026](https://web.archive.org/web/20260802131602/https://cohere.com/pricing) listed Embed 4 Small at $4.00 an hour ($2,500 a month) and every Rerank 4 Medium row at $5.00 an hour ($3,250 a month). The Small embedding tier and the rerank rows above are $1.00 an hour lower; the Medium embedding tier kept its $5.00 rate.

## Is the Cohere API free?

Yes, for testing. Every account gets a trial API key automatically, and calls made with it are free. The limits are tight: Cohere's [rate-limit docs](https://docs.cohere.com/docs/rate-limits) cap trial keys at 1,000 API calls a month and 20 chat requests a minute, and the pricing page forbids using trial keys for production or commercial work.

A production key comes from the Billing and Usage page in the dashboard, after you accept Cohere's SaaS agreement in the [go-live flow](https://docs.cohere.com/docs/going-live). Production calls are billed pay-as-you-go, and **Cohere issues the bill at the end of each calendar month or as soon as you owe $250**, whichever comes first. There is no monthly subscription for the API itself.

## What are the hidden costs?

- **Sales-only flagship.** Command A+ and the other Command A variants have no listed production rate, so your best model's cost is unknown until sales replies.
- **Self-hosting the free weights is not free.** Command A+ runs on two H100 GPUs per Cohere, and that hardware or cloud bill is yours.
- **Image embeddings may cost more.** Embed 4 charged $0.47 per 1M image tokens against $0.12 for text in archived copies; Embed 5 lists a single rate, so confirm how images bill before a large multimodal job.
- **Rerank chunking.** Under the search definition still in the FAQ, any document over 500 tokens is split, and each chunk counts as a document. Long documents inflate rerank volume.
- **Vault hourly vs monthly.** An always-on instance on hourly billing costs $190 more per month on the Small tier than the monthly rate (our arithmetic).
- **Platform pricing.** North and Compass show "Get in touch for custom enterprise pricing" and nothing else.

## Is Cohere worth it in 2026?

Cohere is worth it for retrieval. Embed 5 and Rerank 5 are cheap, priced per token and published, and they plug into any stack, including one that generates answers with another vendor's model. If you are building search or RAG, start there and compare against the stores and frameworks in our [best vector databases](https://dupple.com/learn/best-vector-databases) and [best RAG tools](https://dupple.com/learn/best-rag-tools) guides.

For generation, it depends on what you need. Command R and R7B are inexpensive and predictable. Command A+ is attractive if you want open weights you can host privately, which is the pitch behind Cohere's sovereign and on-prem deployments. If you want a frontier model at a public per-token price with no sales call, Cohere is no longer the obvious choice, and a router from our [best LLM gateways](https://dupple.com/learn/best-llm-gateways) list lets you mix Cohere retrieval with another provider's chat model.

Teams buying an agent platform rather than an API should weigh North against the options in our [best enterprise AI agents](https://dupple.com/learn/best-enterprise-ai-agents) and [best AI agent platforms](https://dupple.com/learn/best-ai-agent-platforms) roundups, most of which publish at least an entry price.

## How we checked these prices

We read [Cohere's pricing page](https://cohere.com/pricing), including each pricing tab and the FAQ, plus the [pricing explainer](https://docs.cohere.com/docs/how-does-cohere-pricing-work), [rate-limit docs](https://docs.cohere.com/docs/rate-limits) and [deprecations page](https://docs.cohere.com/docs/deprecations) on 30 October 2026. Past prices come from Internet Archive copies of Cohere's own pricing page, dated in the links. Scenario volumes, the 730-hour month and every total marked as ours are our own arithmetic from list prices. A summary of the plans also sits on [Cohere's Toolradar pricing profile](https://toolradar.com/tools/cohere/pricing).

## FAQ

**How much does the Cohere API cost in 2026?**

Cohere bills per million tokens on a production key. Command R costs $0.15 input and $0.60 output per 1M tokens, Command R7B $0.0375 and $0.15, Embed 5 from $0.08 per 1M and Rerank 5 from $0.02 per 1M, as of 30 October 2026. Command A+ has no listed production rate.

**What is Cohere's cheapest model?**

For chat, Command R7B at $0.0375 per 1M input tokens and $0.15 per 1M output tokens. For retrieval, Rerank 5 Fast at $0.02 per 1M tokens is the lowest rate on the page.

**Is Command A+ free?**

It is free to try on a trial key and free to download under Apache 2.0. Production API use is quote-only: Cohere routes it to sales, and running the weights yourself means paying for your own GPUs.

**Is Command R+ still $2.50 per million tokens?**

Command R+ 08-2024 is listed at $2.50 input and $10 output per 1M tokens, but only as a legacy rate for existing customers. New projects should not budget on it.

**How is Cohere Rerank priced now?**

Rerank 5 bills per token: $0.02 per 1M on Fast and $0.05 on Pro. Rerank 4 billed per 1,000 searches until October 2026, so small searches now cost much less.

**When does Cohere bill you?**

At the end of each calendar month, or earlier once your outstanding balance reaches $250. Trial key calls are free but capped at 1,000 a month.

## Bottom line

Cohere's published prices now cover its retrieval models and its older Command R line, and they are low: a full support-bot stack can run under $700 a month on the cheaper options (our arithmetic). The flagship Command A+ and the North platform are sales-only. Budget retrieval from the table above, and get a written quote before you plan around Command A+.

**Cite this:** Dupple, "Cohere Pricing 2026: Command, Embed and Rerank Costs, and What Is Now Quote-Only", November 2026.

**Pricing out the AI stack for your team?** [Start the Dupple X trial](https://checkout.dupple.com/checkout/dupple-x-yearly-trial/).

Some offers on this page may be paid placements or contain affiliate links.

Related Articles

Article

### Windsurf Pricing 2026: What It Costs Now That It Is Devin Desktop

Windsurf pricing in 2026, now sold as Devin Desktop: Free, Pro $20/mo, Max $200/mo, Teams $40 per full seat with an $80 minimum, and a 5-developer budget.

Article

### 11 Best Vendor Management Software Tools in 2026

Vendor management in 2026: a priced file beside the purchase order, W-9 payments with no seat fee, and quote-only suites for third parties. Checked Sep 24, 2026.

Article

### 10 Best PR Software Tools in 2026

Best PR software in 2026: Muck Rack is quote-only, Semrush AI PR Toolkit Base is $149/mo, and Prezly Essential is $120/mo billed annually.

Article

### Bolt.new Pricing 2026: Free, Pro and Teams, and What Tokens Really Cost

Bolt.new pricing in 2026: Free with 1M tokens and a 300K daily cap, Pro at $25 a month for 10M tokens, Teams at $30 per member, plus yearly prices and token math.

Article

### AssemblyAI Pricing 2026: Universal Models per Hour and Add-On Costs

AssemblyAI pricing in 2026: Universal-2 at $0.15/hr, Universal-3.5 Pro at $0.21/hr, realtime from $0.15/hr, plus add-ons. What 1,000 hours really costs.

Article

### OpenRouter Pricing 2026: Fees, Free Models and Real Cost

OpenRouter bills models at provider list price and takes 5.5% when you buy credits (8% on Business). Free models, BYOK, minimum fees and real costs.
