Parse a complex multi-column PDF with tables into markdown using the LlamaCloud Parse v2 API, with upload, tier selection, job polling and result retrieval

Documented steps

1. Create an API key in the LlamaCloud dashboard and set LLAMA_CLOUD_API_KEY (format llx-...). All requests use header 'Authorization: Bearer $LLAMA_CLOUD_API_KEY' against base URL https://api.cloud.llamaindex.ai.
2. Optionally install an SDK (Python llama-cloud, TypeScript @llamaindex/llama-cloud), or call REST directly.
3. Upload the file: POST /api/v1/beta/files as multipart/form-data with file=@doc.pdf and purpose=parse. The response returns a file id.
4. Create the parse job: POST /api/v2/parse with JSON body {"file_id":"<id>","tier":"agentic","version":"latest"}. The response returns a job with id and status PENDING.
5. Choose the tier deliberately: fast (1 credit/page) for plain-text bulk documents, cost_effective (3 credits) for text-heavy documents with simple tables, agentic (10 credits) as the default for real-world PDFs with tables, scans and multi-column layout, agentic_plus (45 credits) for dense financial or scientific documents with equations and charts. Tier guide: https://developers.llamaindex.ai/llamaparse/parse/guides/tiers/
6. Poll GET /api/v2/parse/{job_id}?expand=markdown until status is COMPLETED, FAILED or CANCELLED.
7. On COMPLETED, read the inline markdown field. Use expand=items instead for structured JSON (tables, headings, figures), or expand=markdown_content_metadata to get a presigned download URL when the output is large.
8. Retrieval patterns: https://developers.llamaindex.ai/llamaparse/parse/guides/retrieving-results/
9. For mixed-complexity batches, enable the Cost Optimizer feature so individual pages are routed to cheaper tiers automatically.
10. Pin a dated version string (for example 2026-01-08) rather than 'latest' whenever reproducible output across re-runs matters.