> ## Documentation Index
> Fetch the complete documentation index at: https://docs.heygarth.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Claude API Tier Limits

> Understand Anthropic's usage tiers so you can choose the right tier for your team's pull request review volume.

Anthropic automatically upgrades your API account through usage tiers as your cumulative credit purchases increase. Each tier unlocks higher rate limits — more requests per minute, more input tokens per minute, and more output tokens per minute.

<Info>
  Rate limits are enforced at the organisation level, measured separately per model. You can monitor your current tier and live usage in [Anthropic Console → Limits](https://console.anthropic.com/settings/limits).
</Info>

***

## Spend tiers

Advancing to the next tier requires a cumulative credit purchase that meets the threshold below. You advance immediately upon reaching it.

| Tier       | Target audience                           | How to qualify                         |
| ---------- | ----------------------------------------- | -------------------------------------- |
| **Start**  | Evaluation and small pilots               | Automatic upon initial credit purchase |
| **Build**  | Growing teams with regular PR activity    | Based on consistent usage and payments |
| **Scale**  | Large organizations with high PR velocity | High sustained usage and spend         |
| **Custom** | Enterprise scale                          | Contact Anthropic sales                |

<Note>
  **Credit purchase** is the cumulative amount deposited into your Anthropic account (excluding tax), not your total spend. You advance as soon as you cross the threshold — there is no waiting period.
</Note>

***

## Rate limits by tier

Rate limits are measured in three dimensions:

* **RPM** — requests per minute
* **ITPM** — uncached input tokens per minute (cached tokens do not count for most models)
* **OTPM** — output tokens per minute

### Start Tier

Entry-level limits for evaluation.

| Model             | RPM | ITPM   | OTPM   |
| ----------------- | --- | ------ | ------ |
| Claude Sonnet 4.x | 50  | 40,000 | 10,000 |
| Claude Haiku 4.5  | 50  | 50,000 | 10,000 |

### Build Tier

Suitable for small to medium teams with moderate PR volume.

| Model             | RPM   | ITPM      | OTPM    |
| ----------------- | ----- | --------- | ------- |
| Claude Sonnet 4.x | 2,000 | 1,000,000 | 200,000 |
| Claude Haiku 4.5  | 2,000 | 2,000,000 | 400,000 |

### Scale Tier

Suitable for large organisations with high PR velocity.

| Model             | RPM   | ITPM      | OTPM      |
| ----------------- | ----- | --------- | --------- |
| Claude Sonnet 4.x | 4,000 | 4,000,000 | 800,000   |
| Claude Haiku 4.5  | 4,000 | 8,000,000 | 1,600,000 |

<Note>
  **Sonnet 4.x** limits apply to combined traffic across all 4.x models. Exact limits scale dynamically based on your usage patterns.
</Note>

***

## How ITPM works (cache-aware)

A key advantage of the Claude API is that **cached input tokens do not count towards your ITPM rate limit** on current models. Only uncached input tokens and tokens being written to cache consume your ITPM quota.

| Token type            | Counts towards ITPM?          |
| --------------------- | ----------------------------- |
| Uncached input tokens | Yes                           |
| Cache creation tokens | Yes                           |
| Cache read tokens     | **No** (current models)       |
| Output tokens         | Counted under OTPM separately |

This means effective throughput is significantly higher than the raw ITPM number suggests. If 80% of your input tokens are served from cache, you can process 5× more total input tokens per minute than your ITPM limit implies.

<Tip>
  Garth uses prompt caching for system instructions and repository context. Teams with many similar PRs (e.g. same monorepo) benefit automatically from high cache hit rates, increasing effective throughput without needing a higher tier.
</Tip>

***

## Choosing the right tier for your team

<Columns cols={1}>
  <Card title="Start Tier — Evaluation or very small teams" horizontal icon="seedling">
    Fewer than 10 developers, infrequent PRs, or you are trialling BYOK before committing. Active immediately upon funding your account.
  </Card>

  <Card title="Build Tier — Small to medium teams" horizontal icon="chart-line">
    10–50 developers with regular PR activity. The higher RPM and ITPM limits handle dozens of concurrent reviews comfortably.
  </Card>

  <Card title="Scale Tier — Large engineering orgs" horizontal icon="rocket">
    50+ developers, monorepos, or CI pipelines generating a high volume of short-lived PRs.
  </Card>
</Columns>

***

## Token consumption per review

Each Garth review consumes input tokens (your diff and context) and output tokens (the review comments). Estimates below are for Claude Sonnet 4.x.

| Review size                           | Typical input tokens | Typical output tokens |
| ------------------------------------- | -------------------- | --------------------- |
| Small PR (1–5 files, \< 200 lines)    | 3,000 – 8,000        | 500 – 1,500           |
| Medium PR (5–20 files, 200–800 lines) | 8,000 – 25,000       | 1,500 – 4,000         |
| Large PR (20+ files, 800+ lines)      | 25,000 – 80,000      | 4,000 – 10,000        |

<Tip>
  Keeping PRs focused (under 400 lines changed) reduces token usage per review and generally improves comment accuracy.
</Tip>

***

## Rate limit errors

If your account hits a rate limit, Garth retries automatically with exponential backoff. Sustained limits (e.g. many large PRs merging simultaneously) may delay review posting.

**What you will see:**

* A delayed review comment once the retry succeeds
* A dashboard notification if retries are exhausted and the review is dropped

**How to resolve:**

* Advance to the next tier by purchasing additional credits in the [Anthropic Console](https://console.anthropic.com/settings/billing)
* Enable or increase prompt caching to reduce ITPM consumption
* Contact [support](mailto:supports@darner.ai) if you need help sizing the right tier

***

## Getting an Anthropic API key

<Steps>
  <Step title="Create an Anthropic account">
    Go to [console.anthropic.com](https://console.anthropic.com) and sign up or log in.
  </Step>

  <Step title="Purchase credits">
    Navigate to **Billing** and purchase at least \$5 in credits. This activates Tier 1 immediately. Purchase \$40 total to reach Tier 2, \$200 for Tier 3, or \$400 for Tier 4.
  </Step>

  <Step title="Generate an API key">
    Go to **Settings → API Keys** in the Anthropic Console and click **Create Key**. Give it a descriptive name such as `garth-byok`.
  </Step>

  <Step title="Copy and store the key">
    Copy the key — it is shown only once. Paste it into Garth's [Settings → Integrations → LLM](https://app.heygarth.ai/settings/integrations).
  </Step>
</Steps>

<Warning>
  Never commit your Anthropic API key to a repository. Garth's secure vault is the correct place to store it. Anthropic's secret scan — and Garth's own code scan — will flag any key found in source code.
</Warning>

***

## Provider references

Official Anthropic documentation for models, pricing, and rate limits.

<CardGroup cols={2}>
  <Card title="Claude models overview" icon="book" href="https://docs.anthropic.com/en/docs/about-claude/models/overview">
    Full list of available Claude models with context windows and capabilities.
  </Card>

  <Card title="API rate limits" icon="gauge-high" href="https://docs.anthropic.com/en/api/rate-limits">
    Official rate limit reference for all tiers and models.
  </Card>

  <Card title="Pricing" icon="receipt" href="https://www.anthropic.com/pricing">
    Per-token pricing for all Claude models.
  </Card>

  <Card title="Anthropic Console — API keys" icon="key" href="https://console.anthropic.com/settings/keys">
    Create and manage your Anthropic API keys.
  </Card>
</CardGroup>

***

## Next steps

<Columns cols={2}>
  <Card title="BYOK overview" icon="key" href="/byok/overview">
    Add your Claude API key to Garth's secure vault and set it as your active provider.
  </Card>

  <Card title="Anthropic Console — Limits" icon="arrow-up-right-from-square" href="https://console.anthropic.com/settings/limits">
    View your current tier and live rate limit usage in the Anthropic Console.
  </Card>
</Columns>