Token-Based Pricing

​

Overview

Augment uses token-based pricing. Instead of buying a pool of credits and learning a separate conversion table, you pay for the actual resources your work consumes. Every request is billed as the sum of three components:

• LLM tokens at the model provider’s public API list price
• Compute for time Cosmos spends running your work, at our published per-hour rate
• A flat 40% service fee on LLM usage, which funds the Context Engine and the Cosmos platform

This gives you the flexibility to choose the right model for each task, with full transparency on what every request costs.

​

How tokens are billed

When you use Cosmos or Auggie CLI, three things can cost money:

• LLM tokens — every model call uses input and output tokens. You’re billed at the model provider’s public API list price.
• Service fee — Augment adds a 40% service fee on LLM usage. There is no service fee on compute. The service fee covers the Context Engine that hosts and dynamically indexes your codebase, and the Cosmos platform that runs your sessions.
• Cosmos compute — when Cosmos runs work in a sandbox, you pay for the compute time it consumes at $0.19 per hour, billed in 5-minute increments (rounded up). A session that uses 7 minutes of compute is billed as 10 minutes; a session that uses 16 minutes is billed as 20 minutes.

All three components draw from the same usage balance.

​

The Business plan

If you need more than 50 seats or custom Cosmos setup, see Pricing Plans for Enterprise options.

​

A typical month on the Business plan

Here’s how a developer might spend their $100 of included usage in a typical month: Your actual mix depends on the models you choose, how much you use Cosmos, and the size of the tasks you run.

​

Top-ups

Once you’ve used your $100 of included usage, billing continues automatically on pay-as-you-go at the same rates: LLM tokens at public API list price, the 40% service fee on LLM, and compute at $0.19/hour. There’s no minimum top-up amount.

​

Token costs by model

Different models consume your usage balance at different rates based on their underlying provider pricing. All rates below are per million tokens, at the provider’s public API list price. GPT model rates are the standard (default) API tier.

​

Prism routed pricing

Prism uses variable pricing because each request is routed instead of billed against a fixed model rate. When you choose a Prism option, Augment selects the best-fit model from that family based on the task, context, and current system conditions, and the cost reflects the route used. Prism is designed to cost, on average, 20–30% less than frontier model costs, with minimal quality tradeoff. Exact savings depend on the task.

​

Example tasks

To illustrate how different models compare, here are examples of tasks and their approximate cost across each model. All costs include the 40% service fee on LLM usage and assume no Cosmos compute time.

​

Sonnet task: Fix a 500 error in an API endpoint

The /api/users/:id endpoint returns 500 errors when a user has no associated organization. Add null checking and return a 404 with a clear error message instead. Test that users with organizations still work correctly.

Cost: ~$0.25 with Sonnet 4.6 (or Sonnet 4.5)

​

Opus task: Design and implement a multi-tenant billing system

Our B2B SaaS platform needs a multi-tenant billing system. Design the database schema, implement the metering service, integrate with Stripe’s usage-based billing API, and handle edge cases like mid-cycle plan changes, prorations, and failed payments. Consider how this will scale to 10k+ tenants.

Cost: ~$0.85 with Opus 4.7 (or Opus 4.6, Opus 4.5)

​

Monitoring token usage

You can track token consumption from Auggie CLI or the web dashboard.

​

In Auggie CLI

Monitor usage per session directly from the CLI as you work.

​

On the web

Visit app.augmentcode.com for detailed dashboards that show:

• Total usage by your team in dollars
• Usage per team member
• Breakdown by model, activity, and component (LLM, service fee, compute)
• Usage trends over time
• Remaining included usage and pay-as-you-go spend

Team administrators can access more detailed analytics to optimize usage and identify opportunities to save by using more efficient models for appropriate tasks.

​

Understanding your usage breakdown

Usage is organized by model, activity type, and component (LLM, service fee, or compute).

​

Session types

​

Optional features

​

Background activities

Augment performs lightweight processing in the background to keep your experience smooth. These background activities use a small fraction of your total token consumption.

​

Tips for optimizing token usage

1. Match the model to the task. Use Haiku for simple tasks, GPT-5.1 for medium tasks, Sonnet for complex work, Opus for the hardest challenges, GPT-5.4 for computer use and multi-agent orchestration, GPT-5.5 for the highest-complexity coding workflows, and Prism when you want Augment to balance quality and cost by routing within a curated model family.
2. Be specific in your prompts. Clear, detailed instructions help models work more efficiently and use fewer output tokens.
3. Lean on prompt caching. Augment automatically caches stable context (your repo index, AGENTS.md, recently used files). Cached input tokens are billed at the provider’s reduced cache rate — typically around 10% of the input rate — and the service fee scales with them.
4. Break down large tasks. Sometimes splitting a complex task into smaller ones is more token-efficient than planning and executing everything in one session.
5. Scope your Cosmos sessions. Compute is billed at $0.19/hour in 5-minute increments with no service fee. Closing a session when you’re done releases the compute and stops the meter at the next increment boundary.
6. Review usage patterns. Check your dashboards regularly to identify optimization opportunities. Team admins can set guidelines on which models to use for which work.

​

Frequently asked questions

​

Related resources