Standard Vault Pricing

Standard Vault is billed as a Cohere-managed service. Pricing depends on the models you select and each model’s performance tier. Cohere manages the underlying infrastructure and scaling, and customers can choose between two pricing models:

Feature	Fixed	Flex
Commitment	Monthly or annual	Monthly or annual
Capacity	Fixed number of instances (no autoscaling)	Minimum baseline instances, plus autoscaling
Sizing	Determined through a sizing exercise or a production trial (for example, based on expected load)	—
Autoscaling	—	Scales up/down based on request rate and agreed latency SLOs
Pause/resume	You can pause a model or restart a paused model to save on costs.	You can pause a model or restart a paused model to save on costs.
Overages	—	Additional capacity billed per instance-hour
Max capacity	—	Maximum instance cap per model

The following table summarizes the available models and their rates. All rates are per instance.

Model	Performance Tier	Hourly rate	Monthly rate	Annual rate
Embed 4	Small	$4.00	$2,500	$25,000
Embed 4	Medium	$5.00	$3,250	$32,500
Rerank 3.5	Medium	$5.00	$3,250	$32,500
Rerank 4 Fast	Medium	$5.00	$3,250	$32,500
Rerank 4 Pro	Medium	$5.00	$3,250	$32,500
Rerank 4 Pro	Large	$10.00	$6,500	$65,000

You may also want to compare Standard Vault pricing against the operational and capacity costs of running inference directly in your cloud provider account (for example, AWS), where cloud-provider credits may apply.

Generative models

The rates above cover Embed and Rerank, which are available self-serve. Generative models (the Command A family and North) are also available on Standard Vault, typically through a waitlist. All rates are per instance, per hour.

Model	Hourly rate	XL hourly rate
Command A	$40.00	$48.00
Command A Vision	$40.00	$48.00
Command A Translate	$40.00	$48.00
Command A Reasoning	$48.00	$57.50
Command A+	—	$57.50
North Large	—	$57.50
North Mini Code	—	$57.50

Monthly and annual commitment pricing follows the same Fixed and Flex plans described above; contact Cohere for those rates and for access. Generative access typically requires a waitlist, so check the model dropdown when creating a vault, or contact Cohere. Bundles and customized models are also available; see Supported Models for the full list.

For the pricing of the encrypted, confidential-computing product, see Model Vault Encrypted Pricing.

Performance Tiers

Each model has a performance tier based on latency requirements and throughput service level objectives (SLOs) per instance. You can see the tiers listed in the model dropdown selection as a size letter (e.g., S, M, L). The tiers follow an instance/hour pricing, which is then incorporated into your payment plan. We recommend selecting the model-performance tier combination that matches the nature of your workflow and the required measures of performance.