# Lune vs Parallel

Parallel builds retrieval infrastructure for agents over the open web, priced per thousand requests. Lune builds the same thing for a much narrower corpus. The real question is which one your answers need.

Competitor site: https://parallel.ai.

## Choose Parallel if

- Your agent needs the live web: company data, docs, filings, news.
- You want accuracy tiers per request and will build the product around them.
- Monitoring pages over time or extracting at scale is the job.

## Choose Lune if

- Your answers must cite published research, because someone will check.
- You need the definition, the ablation table, the stated limitation.
- You would rather install an MCP server than integrate a REST API.

## From a question to a citation

Parallel: Query the open web -> Extract from pages -> Structured result.
Lune: Query top-tier venues -> Read the full text -> Quote with a locator.

Same idea, different corpus. Most serious agents want both.

## Line by line

| Dimension | Lune | Parallel |
| --- | --- | --- |
| The corpus | Peer-reviewed papers from top-tier venues, in full text. | The open web. |
| What quality means | Accepted at a venue the field ranks top-tier, before indexing. | Accuracy tiers you select per request. |
| How you connect | An MCP server and a CLI, with OAuth or a token. | REST APIs you integrate yourself. |
| What the tools return | Quotes, locators, citation edges, typed tables. | Task, Search, Extract, Monitor and Find All results. |
| Methodology guidance | A corpus on peer review, ablations and venue choice. | Not offered. It is general infrastructure. |
| How it is priced | Flat monthly plans with a daily allowance, from $0. | Per thousand requests, from $1, with 5,000 free a month. |

## Same idea, different corpus

Parallel and Lune agree on something most of this category does not: the useful product for the agent era is retrieval infrastructure, not another chat window.

They disagree on what an agent should read. Parallel says the live web, which is right for almost every commercial question. Lune says the peer-reviewed record, which is right when the claim has to survive a paper, a grant or a safety review.

## What full text buys

Web retrieval returns pages. For a paper that is usually an abstract, a landing page, or a PDF someone still has to parse, and the parse is where numbers, tables and equations go missing.

Lune parses once, at ingest, and keeps the structure. Doing that work before the agent asks is why the tools answer quickly.

## Pricing

- Lune Free: $0, 10 requests a day, every tool
- Lune Pro: $4.99, a month or $47.90 a year, 300 requests a day
- Lune Max: $9.99, a month or $95.90 a year, 600 requests a day
- Lune Credits: $0.99, 300 extra requests, no expiry
- Parallel Free: $0, 5,000 requests a month
- Parallel Search API: $1 to $5, per 1,000 requests
- Parallel Task API: $5 to $2,400, per 1,000, by tier
- Parallel Extract API: $1, per 1,000 results

Parallel lists several more APIs with their own rates.

## Questions

### Is Lune a Parallel alternative?

Only if your agent's questions are scientific. Parallel is web infrastructure; Lune is literature infrastructure. Most research agents want both.

### Does Parallel search academic papers?

Its APIs are general web retrieval and do not advertise an academic index. What it finds about a paper is what the web says about it.

### Does Lune have a REST API?

The supported paths are the MCP server and the CLI. The underlying service contract is documented but is not a public integration interface.

## What we read

- [parallel.ai](https://parallel.ai)
- [Pricing](https://parallel.ai/pricing)