Tiers

Parse offers four tiers that trade off cost, latency, and accuracy. You pick a tier on every parse request, plus a version so the behavior stays reproducible. This page is the canonical reference for what each tier does, when to pick it, and how the versioning model works.

Which tier should I pick?

Use this as a starting point. You can always re-parse the same file with a different tier — there’s no setup cost to switching.

My document looks like…	Pick
Plain text, single column, no tables — and I just need raw text or layout-preserving spatial text	Fast
Mostly text with simple tables, need clean markdown, want to keep cost down	Cost Effective
Scanned pages, multi-column layouts, real tables, embedded charts	Agentic
Dense financial reports, complex tables, charts and diagrams, mission-critical accuracy	Agentic Plus
Mixed — most pages are simple but a few are complex	Agentic or Agentic Plus with Cost Optimizer enabled

Mixed-complexity documents: use Cost Optimizer

If your document has a mix of simple prose pages and a few complex pages with tables or charts, the most cost-effective choice is usually not picking a single tier — it’s enabling Cost Optimizer on top of agentic or agentic_plus. Cost Optimizer routes each page to the right tier automatically and runs both groups in parallel, so you get premium quality where it matters and cost_effective prices on the easy pages — without latency penalty.

Credit cost per page

Credit costs increase with tier capability — Fast is the cheapest and Agentic Plus is the most expensive. See the Pricing page for current rates, volume discounts, and pricing for other Parse output options (e.g. layout extraction).

Caching is free. Re-parsing the same file within 48 hours costs zero credits — Parse caches results by default. See Cache Control for how to opt out of cache when you need a fresh parse.

Versioning and reproducibility

Tiers in Parse v2 are versioned. When you specify a tier with a specific dated version like "agentic" with version="2026-01-08", the parsing behavior is locked: the same model, the same prompts, the same configuration, every time. This is the contract that makes Parse safe to use in production — your output won’t shift under you because we shipped a new model.

Use "version": "latest" during development to always get the most recent stable release. For production, pin a published version date. Only dates that correspond to an actual release are accepted — arbitrary dates will return a validation error.

{
  "tier": "agentic",
  "version": "2026-01-08"
}

When a newer version ships, you decide when (and whether) to bump. We don’t break existing pinned versions — they stay available for the foreseeable future, so you can upgrade on your own schedule and roll back if a new version doesn’t suit your workload.

tier and version are both required on every parse request. Forgetting either will return a validation error.

Available tiers

Each tier is a snapshot of a particular model + prompt + configuration combo. Tiers don’t share models — picking agentic_plus doesn’t just turn on extra options on top of agentic, it’s a different parsing pipeline.

For complete API request examples, see Getting Started or the REST API Guide.

Agentic Plus

Best for: the hardest documents — dense financial reports, scientific papers with equations and figures, complex multi-column layouts, and anything where accuracy is mission-critical.

Notable defaults: Specialized chart parsing is on by default at this tier — no extra configuration needed.

{
  "tier": "agentic_plus",
  "version": "latest"
}

With explicit version pinning for production:

{
  "tier": "agentic_plus",
  "version": "2026-01-08"
}

Agentic

Best for: complex but not extreme documents — most real-world PDFs with tables, scanned pages, multi-column layouts, charts and diagrams. This is the default recommendation for most use cases that need high-quality structured output.

Pair with Cost Optimizer on long mixed-complexity documents to drop the per-page cost on simple pages.

{
  "tier": "agentic",
  "version": "latest"
}

Cost Effective

Best for: text-heavy documents with simple tables and minimal visual structure — content where you still want clean markdown and table preservation, but can trade off the deepest layout reasoning for a cheaper run.

{
  "tier": "cost_effective",
  "version": "latest"
}

Fast

Best for: plain-text documents at high volume where you only need text and layout-preserving spatial text. The fastest tier, with the lowest per-page cost.

Important limitation: the fast tier does not support markdown output. Requesting expand=["markdown"] on a fast-tier job returns a validation error. If you need markdown, structured items, or chart data, use Cost Effective or higher. Fast is also the only tier that does not support agentic_options (custom prompts and other agentic-model controls), since it doesn’t run an agentic model.

{
  "tier": "fast",
  "version": "latest"
}

Example API request

Here’s a complete request showing the tier + version pattern:

Python

from llama_cloud import LlamaCloud

client = LlamaCloud(api_key="llx-...")

result = client.parsing.parse(
    upload_file="example_file.pdf",
    tier="agentic_plus",
    version="latest"
)

TypeScript

import fs from "fs";
import { LlamaCloud } from "@llamaindex/llama-cloud";

const client = new LlamaCloud({ apiKey: "llx-..." });

const result = await client.parsing.parse({
  upload_file: fs.createReadStream("example_file.pdf"),
  tier: "agentic_plus",
  version: "latest"
});

cURL

curl -X 'POST' \
  'https://api.cloud.llamaindex.ai/api/v2/parse' \
  -H 'Accept: application/json' \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $LLAMA_CLOUD_API_KEY" \
  --data '{
    "file_id": "<file_id>",
    "tier": "agentic_plus",
    "version": "latest"
  }'

What’s new

Parse uses dated tier versions — a release is typically a new dated version of one of the four tiers. Pinned versions stay available so you can upgrade on your own schedule.

The Parse team ships new tier versions roughly monthly. The source of truth for what "latest" resolves to is the API itself — make a parse request with "version": "latest" and inspect metadata.version on the response.

Notable feature releases

• Cost Optimizer — per-page tier routing on agentic or agentic_plus. See Cost Optimizer.
• Agent Skill — native document parsing for coding agents. See Getting Started → Agent Skill.
• Chart parsing default-on at Agentic Plus — specialized_chart_parsing is enabled by default on agentic_plus. See Specialized Chart Parsing.
• v2 API — replaced v1’s flat 70+ form parameters with structured JSON. tier and version are now required. See the REST API Guide.

See also

• Cost Optimizer — the per-page tier router for mixed-complexity documents
• Pricing — canonical credit costs and volume discounts
• Configuring Parse — where every option lives in the request shape
• Getting Started — install the SDK and run your first parse
• Specialized Chart Parsing — default-on with Agentic Plus, opt-in elsewhere