Cohere Model Rates, Rerank Units & the Real Bill 2026 Guide
Cohere charges per model and per endpoint, not per seat, so the plan name barely matters. The lever that sets most bills is Rerank, priced by search unit rather than by token. Here is the real math.
Typical token rate
$0.0375-$10/1M
Command R7B input to Command A output; usage-based, no subscription
Hidden fees
Yes
Rerank billed by search unit, dedicated Model Vault, image embeddings priced apart
Free tier
Trial key
rate-limited access to every endpoint for non-production testing
Cost transparency
Medium
scores 4 of 6 on our transparency checklist
Cohere true cost, per model and endpoint
High· Verified July 15, 2026Cohere is usage-priced rather than a flat subscription as of July 15, 2026, with a free trial key to start. Command R7B is the cheap model at $0.0375 per million input, while the flagship Command A runs $2.50 in and $10 out. Rerank bills by search unit at about $2 per 1,000, and Embed and dedicated Model Vault price on their own scale. A production key bills at month end or once you cross $250 outstanding. So the model and endpoint you call, not a plan, decide the bill.
- Command R7B in /1M$0.0375
- Command R in /1M$0.15
- Command A in /1M$2.50
- Command A out /1M$10
- Rerank /1k searches$2
- Embed v3 /1M$0.10
- Embed 4 images /1M$0.47
Cohere's cheapest model, Command R7B, starts at $0.0375 per million input, far below the $7.99 median across the 20 llm tools we track. The flagship Command A runs many times that.
What the free Cohere trial key gets you
Cohere's free tier is a trial key, not a free plan with quotas. It gives rate-limited access to every endpoint, generate, embed and rerank, for non-production testing at $0. That is exactly enough to build a pipeline and prove it works before a cent changes hands, which is the point.
The limit is production. The trial key is throttled and not meant for live traffic, so the moment you ship, a production key meters you per model and per endpoint. Prove the retrieval quality on the trial, then move to production billing with your model routing already decided. If Cohere's search focus is more than your use case needs, the Cohere alternatives page shows where a general model API might cost less.
Cohere savings that come from the endpoint you pick
Cohere has no seats and no coupons. There is no student, nonprofit or academic rate as of July 2026, because a per-model API leaves a discount nothing to sit against. The savings are structural, and on a retrieval workload they are large.
Route generate calls to Command R7B wherever quality allows, since it costs a fraction of Command A. Rerank only the candidates that matter, because chunking multiplies search units. Keep dedicated Model Vault for steady volume rather than spiky traffic. Above real usage, Enterprise and on-prem deployment are quote-based, and that negotiated lane is where the tactics below apply.
Route to Command R7B
At $0.0375 per million input against Command A's $2.50, the small model is aggressively cheap for high-volume, low-latency calls. Reserving the flagship for genuine reasoning cuts a generate bill by orders of magnitude.
Trim what you rerank
Because Rerank bills by search unit and chunks long documents, reranking only the top candidates rather than the full set keeps the count down. Disciplined retrieval upstream is a direct saving on the Rerank meter.
Enterprise and on-prem
For regulated buyers, private-cloud or on-prem deployment with custom models is quote-based, which makes the list an anchor. Committed volume and a term open room the self-serve card never shows.
Match Model Vault to steady load
A dedicated deployment bills hourly whether busy or not, so it only pays off at consistent high throughput. For spiky traffic, the shared per-token API is usually cheaper than reserving capacity you cannot keep full.
Negotiating a Cohere Enterprise deployment
The per-model card does not move, and Rerank and Model Vault rates are fixed for everyone on the shared API. The levers there are engineering: model routing and disciplined reranking. Real negotiation opens at Enterprise, where private deployment, custom models and committed volume are quote-based.
Two levers do most of the work here, and both rest on Cohere valuing a committed deployment over metered spot calls.
Commit search and token volume
- Target
- Enterprise contract
- Argument
- Guarantee a monthly generate, embed and rerank volume for rates under the public card. Reserved retrieval revenue is worth a discount to Cohere, and a forecastable bill matters more here than on a chat API because Rerank is hard to predict.
Anchor on a general model API
- Target
- Any high-volume deal
- Argument
- If your pipeline does not strictly need Cohere's retrieval tuning, Amazon Nova near $0.035 per million or Mistral Large at $2 in are cheaper. Make Cohere defend the premium on rerank quality, or match the rate.
Bundle Model Vault into the term
- Target
- Dedicated deployment
- Argument
- If you need reserved capacity, fold the Model Vault hours into the committed contract rather than paying the $4-to-$10 hourly card. Reserved, forecastable capacity is exactly what a negotiated rate rewards.
Timing a Cohere Enterprise deal
The shared API rate card has no timing angle, because it does not track the calendar. Only an Enterprise or on-prem conversation is worth timing, and it follows the standard sales cycle. Cohere's sales team carries quarterly numbers, so a committed deployment you can sign as a quarter closes tends to price sharper than one opened in its first weeks.
Jan
Feb
Mar
Q-END
Apr
May
Jun
Q-END
Jul
Aug
Sep
Q-END
Oct
Nov
Dec
Q-END
Pro tip: Bring an audit of your real rerank and token volumes, not a projection. Rerank in particular is hard to forecast, so a measured usage history is what earns a rate below the shared card and keeps the deployment sized right.
Cohere costs: what bends, what holds firm
Aim requests where Cohere can move. The shared per-model card and Rerank rates hold for everyone; the room is in committed volume and enterprise deployment.
Usually negotiable
- Committed per-model token rateHIGH
- Rerank and search-unit pricing at volumeHIGH
- Model Vault dedicated termsMEDIUM
- On-prem and private-cloud deploymentMEDIUM
- Payment terms (Net 30/60)LOW
Rarely negotiable
- Published per-model token rates
- The shared Rerank per-search rate
- Embed and image-embedding rates below volume
- The $250 billing-trigger threshold
Cohere negotiation email generator
Enter your details and the draft below writes the copy, quoting live rival rates from our catalog. Route it to your Cohere account contact or the sales form. The order carries the weight. State your monthly token and search volume, place a competing rate alongside, ask for a committed rate or an on-prem quote, and give a date to close by.
private cloud or on-prem, custom models, committed volume
Hi Cohere team, I lead tooling decisions at [Your company], and we are evaluating an enterprise credit pool for our team of 10-50 people. As part of this evaluation we are also looking at Amazon Nova, which comes in at $0.035 per 1M input, and Mistral Large at $2 per 1M input. Can you help us understand the value difference at your current rates? We are ready to commit to an annual term. What is the best rate you can offer on annual billing, and can you cap the renewal price in the contract? We are aiming to sign before the end of this quarter, and budget sign-off is already in place. Could you share a proposal covering the per-seat or per-credit rate, the renewal terms, and any programs we qualify for? Best regards, [Your name] [Your company]
Send it Tuesday to Thursday, and follow up once after 3 business days.
Before you send
- Bring your monthly generate, embed and rerank volumes. A commitment ask with no numbers behind it stalls.
- Send midweek, since a note landing Tuesday through Thursday moves faster than a Monday or Friday one.
- Keep your ceiling private and let Cohere quote the committed rate first, then push against it.
- Name two rival model APIs in the note. The generator inserts their current token rates for you.
- Get the rerank rate, the Model Vault terms and the commitment in writing before you shift production.
- Chase once after a few business days, then read continued quiet as a read on your position.
Cohere cost mistakes in RAG pipelines
Each of these grows out of how Cohere prices search and retrieval, and all are avoidable early.
Reranking the whole candidate set. Rerank bills by search unit and chunks long docs, so trim to the top results.
Running generate on Command A by default. Command R7B is a fraction of the rate for most retrieval answers.
Reserving Model Vault for spiky traffic. It bills hourly whether busy or not, so it only pays at steady load.
Forgetting image embeddings price apart. Embed 4 images are $0.47 per million on top of text embedding.
Missing the $250 billing trigger. A production key bills at month end or once you cross that outstanding balance.
Cohere rivals to weigh for search and RAG
Cohere is tuned for retrieval, so leverage means naming the general APIs that could serve your pipeline for less. These three post rate cards you can lay beside Cohere's, checked against our catalog. Test one on your own recall and rerank quality, and the committed-rate conversation rests on evidence rather than Cohere's own framing of the market.
Amazon Nova
Nova Micro input, batch cheaper
$0.035/1M
The price floor for general inference. If your pipeline does not need Cohere's rerank tuning, Nova undercuts even Command R7B on plain generation.
Mistral Large
input, EU-hosted, batch 50% off
$2/1M
A capable general model with European hosting. The card for regulated retrieval teams who want an alternative on data residency, beyond rate alone.
Google Gemini
input rate, grounding available
$1.25/1M
Native Search grounding is a different route to fresh answers than a rerank pipeline. Worth pricing when grounding could replace part of your retrieval stack.
Script“For our RAG workload, Amazon Nova is $0.035 per million and Mistral Large $2 in. What committed rate can Cohere put on generate and rerank to keep us here?”
Is Cohere worth it for retrieval? A cost read
Cohere is built for search and RAG, and priced for it, which is both the strength and the thing to watch. Command R7B is genuinely cheap for high-volume retrieval, the models are solid, and the free trial key lets you prove a pipeline for nothing. The cost that surprises people is not generate at all. It is Rerank, priced by search unit and multiplied by chunking, plus dedicated Model Vault that bills while idle.
So manage the endpoints, not the model alone. Route generate to Command R7B, rerank only what matters, keep Model Vault for steady load, and treat image embeddings as their own line. Above real volume, take an Enterprise conversation to committed usage with a general API's rate in hand.
Handled that way, Cohere is strong value for retrieval-heavy work. The per-model and per-endpoint rates sit on the Cohere pricing page. This page is here to keep the Rerank line from quietly running the bill.
Cohere pricing and discount FAQ
How does Cohere charge for usage?
+
Cohere is usage-priced per model and per endpoint, with no subscription. Command R7B is $0.0375 per million input tokens, Command R is $0.15, and the flagship Command A is $2.50 in and $10 out. Rerank bills by search unit at about $2 per 1,000 searches, Embed prices on its own scale, and dedicated Model Vault runs $4 to $10 an hour. A free trial key covers testing. A production key bills at month end or once you cross $250 outstanding.
Why is Cohere Rerank more expensive than expected?
+
Because Rerank bills by search unit, not by token, and it chunks long documents. Each document over 500 tokens splits into pieces that each count as a separate document in the search unit. So a pipeline reranking long passages can rack up far more search units than it has queries. At about $2 to $2.50 per 1,000, that adds up quickly. The fix is to rerank only the top candidates and keep passages short, which controls the search-unit count directly.
What does the free Cohere trial key include?
+
It is a rate-limited key giving access to every endpoint, generate, embed and rerank, for non-production testing at no cost. That is enough to build and validate a retrieval pipeline before you pay. It is not meant for live traffic, though, so the throttling will pinch a real workload. Once you ship, a production key meters per model and per endpoint. Use the trial to lock in your model routing and rerank discipline, then move to production billing.
Does Cohere have student or nonprofit rates?
+
No published program as of July 2026, because a per-model, per-endpoint API has no plan to hang a discount on. The savings are structural: route generate calls to the cheap Command R7B, rerank only what matters to control search units, and reserve dedicated Model Vault for steady load. For organizations, Enterprise and on-prem deployment with custom models are quote-based, which is where committed volume actually negotiates a lower rate.
Is Cohere cheaper than a general model API for RAG?
+
It depends on whether you need its retrieval tuning. Command R7B at $0.0375 per million is competitive for generation, and Rerank is a genuinely strong reranker. But a general API like Amazon Nova near $0.035 or Mistral Large at $2 in can be cheaper if your pipeline does not lean on Cohere's rerank quality. Benchmark recall and rerank precision on your own data, because the retrieval quality, more than the token rate, is what justifies staying with Cohere.
How much does Cohere Model Vault cost?
+
Model Vault is dedicated deployment, priced at roughly $4 to $10 an hour, or about $2,500 to $6,500 a month depending on model size. It bills continuously while provisioned, whether or not it serves traffic, because you are reserving capacity rather than paying per call. That makes it worthwhile only at steady, high throughput. For spiky or low-volume workloads, the shared per-token API is usually cheaper than keeping a dedicated deployment warm.
How do you get a lower rate on Cohere?
+
Below committed volume the per-model card holds, so real savings come from engineering. Route generate to Command R7B, rerank only the top candidates, and size Model Vault to steady load. Once your usage is large enough, negotiate an Enterprise or on-prem contract with committed generate, embed and rerank volumes. Bring a general API's rate from Nova or Mistral to anchor, and put the rerank and Model Vault terms in writing. Expect 10 to 25 percent at real scale.
What is the cheapest way to run a retrieval pipeline on Cohere?
+
Route generation to Command R7B at $0.0375 per million, and reserve Command A for answers that truly need the flagship. Rerank only your top candidates and keep passages under 500 tokens so chunking does not multiply search units. Embed text and images as separate meters, and use the shared API rather than a dedicated Model Vault unless your load is steady. Those habits keep the Rerank line, which drives most retrieval bills, firmly under control.
Explore Cohere
Every page on Cohere in one place, you are on cost guide.
Snapshot, score and verdict
How to get API access, limits, SDKs and what it costs
Every tier and the entry price
You are here
Compared and ranked vs peers
Price and feature change history
Browse the full Large Language Models category
Sources & verification
| Source | What was checked | Last checked |
|---|---|---|
| Cohere official pricing | Verified plan prices, renewal rates and credit allowances | July 15, 2026 |
| Cohere website | Official vendor website | July 15, 2026 |
| Cohere pricing on ComparEdge | Current prices for every plan, with the cost calculator | July 15, 2026 |
Every fact on this Cohere pricing page is tied to a named source and a verification date. Freshness-sensitive figures trace to the sources above; verify against the vendor before relying on them.