Compare
LlamaExtract alternative: LlamaExtract and Velrim on price, confidence scores and published evidence
LlamaExtract is the schema-driven extraction product inside LlamaCloud, the hosted platform from LlamaIndex, with $27.5M raised. Velrim is one extraction API with a confidence score on every field and the error curve for that score published. This page sets the two side by side on what each one publishes. Every LlamaExtract entry is a quotation from a LlamaIndex page with the day it was read. Velrim has not run LlamaExtract, because LlamaIndex's terms prohibit it. The clause is quoted below.
Side by side
| LlamaExtract | Velrim | |
|---|---|---|
| What it is | Schema-driven extraction on LlamaCloud, beside LlamaParse, LlamaClassify and LlamaSplit. "Extract structured data from documents using custom schemas." | One extraction API. Schema in, typed fields out, each with a page location and a 0 to 1 confidence score. |
| Extraction list price | Credits at $1.25 per 1,000, extract tier plus parse tier. Cost-effective 8 credits a page ($10.00 per 1,000 pages), Agentic 25 ($31.25), Agentic Plus 60 ($75.00), Turbo 35 ($43.75). | $0.02 per successfully extracted page, $20 per 1,000. Grounding and confidence included, no surcharges. |
| Free tier | Free plan with 10K credits: 1,250 pages on Cost-effective, 400 on Agentic. | ~500 pages free, no card, 90 days. |
| How to buy | Self-serve: Starter $50 a month for 40K credits, Pro $500 a month for 400K. Enterprise: "Custom". | Self-serve. Prepaid balance from $5. |
| Per-field confidence | With confidence_scores: true: parsing_confidence, extraction_confidence and a combined confidence. "Citations and confidence scores will significantly slow down extraction processing time." | One 0 to 1 score per field, always on. |
| What the vendor says the score means | On Cost Effective, Agentic and Agentic Plus, "a score approximates a real probability of correctness". "At a 0.8 threshold roughly 75% of extraction errors fall below the line." Agentic Max and Turbo: "the calibration above does not apply to them." | A probability that the field is correct, with the gap between claimed and observed published: 13 points on average in the published comparison, refusals counted as a coin flip. |
| Evidence for the score | None published on the extensions, configuring or product pages as of 2026-09-04: no dataset, sample size or error curve behind the 0.8 threshold sentence. | Error curves published and regenerated on every refit, and a per-bucket table in the published comparison. |
| Grounding | With cite_sources: true: "The page number where the information was found. The verbatim text that was used to extract the field value." plus a box x, y, w, h and page dimensions. On Turbo, "citations are text-level and carry no bounding boxes." | Page, bounding box and matched snippet on every field, from a word map, no second model call. |
| Benchmarks published | ParseBench, April 2026: ~2,000 pages, 167,000 test rules, 14 methods, LlamaParse Agentic 84.9% overall. A parsing benchmark. | A pre-registered comparison of six extraction setups including Velrim on 124 public documents, September 2026. Every raw output and the scorer are public, DOI 10.5281/zenodo.22233430. |
| Terms on benchmarking | "for benchmarking or competitive analysis of the Service" is a prohibited use, §2.2(h). Terms last updated Jun 7, 2024. | None. The eval CLI exists to be run against Velrim, and ships a LlamaExtract adapter. |
| SDKs | Python, TypeScript, Go and Java SDKs, a CLI, a REST API and a web UI with a schema builder. | TypeScript and Python, MCP server. |
| Funding | $27.5M total, $19M Series A led by Norwest (Mar 2025). | Self-funded. |
What LlamaExtract says about its confidence score
This feature adds three confidence-related fields to the extraction metadata:parsing_confidence: Confidence score indicating how well the relevant context was parsed from the source document.extraction_confidence: Confidence score indicating the relevance of the extraction based on the JSON schema field.confidence: Combined confidence score that incorporates both parsing and extraction confidence.
Reading the scores. confidence is the value to threshold on; the other two explain where a low score came from. Calibrated on Cost Effective, Agentic, and Agentic Plus. On those tiers a score approximates a real probability of correctness, so you can set a threshold directly rather than only ranking fields against each other. At a 0.8 threshold roughly 75% of extraction errors fall below the line.
Agentic Max and Turbo return scores from an earlier model. They are still useful for ranking fields, but the calibration above does not apply to them.
Validate the threshold on your own documents. The right cutoff depends on your document mix and on how costly a missed error is. Start at 0.8, score a sample you have ground truth for, and move it until review volume and escape rate sit where you want them.
Citations and confidence scores will significantly slow down extraction processing time. Enable these features only when the additional metadata is essential for your use case.
The score is three numbers, returned when confidence_scores is on, and the docs name the combined one as the value to threshold. The threshold sentence carries one figure, 75% of errors under 0.8, for three of the five tiers. As of 2026-09-04, the extensions, configuring and product pages give no dataset, sample size or error curve behind it, and the sentence after it asks the reader to score a sample of their own.
What Velrim publishes about its score
One number per field, 0 to 1. In Velrim's published comparison of six extraction setups (September 2026), across every field the score graded, the claimed confidence and the observed share correct differed by 13 points on average with refusals counted as a coin flip, 10 points without them. On the 33 values Velrim itself invented, the score averaged 0.40 against 0.64 overall. The curves are on the error curves page and regenerate on every refit. The definition of the fabrication measure is on the metric page.
Why LlamaExtract is not in the comparison
Except to the extent a restriction is prohibited by Applicable Law, you will not do, and will not assist, permit, or enable any third party to do, any of the following:
(h) use or display the Service in competition with us, to develop competing products or services, for benchmarking or competitive analysis of the Service, or otherwise to our detriment or disadvantage;
The terms carry "Last Updated: Jun 7, 2024" (LlamaIndex Terms of Service, read 2026-09-04). The clause is use-based and the preamble extends it to third parties, so a paid account does not cure it and running through someone else does not either. Velrim sent LlamaIndex a written consent request on 2026-07-12 describing the protocol, the metrics, pre-publication replication access and a standing right of reply, with a deadline of 2026-07-20. No reply arrived by the deadline, and the pre-registered outcome for silence was to drop the arm and disclose. The full record is in the analysis plan. The LlamaExtract adapter in velrim-eval is published and runs under the caller's own account and terms.
What the comparison measured
| Arm | macro-F1 (norm) | fabrication (absent fields) | wrong among the top 90% by confidence | $/1k pages |
|---|---|---|---|---|
| Velrim | 0.691 | 11.5% [6.1, 19.8] | 36.1% [29.9, 44.1] | $20 (list; measured matches) |
| Gemini free-decode | 0.709 | 17.0% [10.1, 26.7] | not requested | ~$2.2 (measured, token-priced) |
| Gemini constrained | 0.697 | 12.8% [7.6, 20.5] | not requested | ~$2.2 (measured, token-priced) |
| gpt-5.4-mini free | 0.735 | 10.8% [5.7, 18.4] | none surfaced | ~$1.0 (measured, prompt-cached) |
| gpt-5.4-mini structured | 0.718 | 10.4% [5.3, 17.6] | none surfaced | ~$1.0 (measured, prompt-cached) |
| Mistral OCR 4 | 0.712 | 40.3% [31.0, 52.5] | none surfaced | $5 (list; measured $5.3) |
Six setups on 124 public documents, 96 of their fields absent by the audited answer key. Pooled, the six invented a value for 17% of those fields. Velrim: 11.5% [6.1, 19.8]. Velrim is the most expensive row and ties on accuracy. The full comparison has the intervals, the per-document-type breakdown and every raw output.
Pricing
| Tier | Credits a page, extract plus parse | Per 1,000 pages at $1.25 per 1,000 credits |
|---|---|---|
| Cost-effective | 8 | $10.00 |
| Agentic | 25 | $31.25 |
| Agentic Plus | 60 | $75.00 |
| Turbo, no parse step | 35 | $43.75 |
| Velrim, extraction with grounding and confidence | $20 |
All features are priced using credits, which are billed per page (or minute for audio). Price per 1,000 Credits: North America $1.25. Europe $1.25.
Extract tier, Extract credits/page, Default parse tier, Parse credits/page, Default total: Agentic Plus, 50, Agentic, 10, 60. Agentic, 15, Agentic, 10, 25. Cost-effective, 5, Cost-effective, 3, 8. Turbo, 35.
Free $0 /month, Includes 10K credits. Starter $50 /month, 40K credits. Pro $500 /month, 400k credits. Enterprise, Custom. 1,000 credits = $1.25.
Turbo bills one all-in per-page rate with no separate parse charge, and that rate is higher than Agentic: it is a premium option that pays for speed, not a cheaper one
A page costs the extract tier plus the parse tier it runs on, so the Agentic rate is 25 credits, or $31.25 per 1,000 pages at $1.25 per 1,000 credits. The free plan's 10K credits are 400 Agentic pages or 1,250 Cost-effective pages. Citations and confidence scores add processing time on every tier, per the docs, and the pricing page lists no extra credits for them. Velrim bills $0.02 per extracted page with grounding and the score included, prepaid from $5.
Which to pick
LlamaExtract, for a schema builder in a web UI, a choice of parsing tiers, and one platform that also parses, splits and classifies, on a free plan with no card. Velrim, when the requirement is a number per field to sort on and the evidence for that number in the open, at one list price with grounding and the score always on.