Rightmark vs LXT: a 2026 head-to-head comparison

Match the asset to the task, then choose who will run the work needed to make it usable.

Check prices, currencies, fee bases and route terms before accepting a proposal. Exact sample tests, acceptance and payment terms remain scope-specific.

Start with the conditions the data must cover and the tests it must pass. Then compare money, rights, approvals, coverage and responsibilities before deciding how much weight to give a provider's track record.

1. Will the data fit the task and pass the tests we agree?

A well-formatted archive can still be the wrong purchase. If a voice assistant must understand a caller over household noise, a collection of quiet office conversations may supply useful vocabulary while leaving the acoustic requirement untouched. Owners need an acceptance description that tells them which part of their archive is being evaluated.

Consider this invented decision: Nora leads speech engineering at the fictional support-software company Hearthline (an example; the company and the person are invented). Assume its assistant needs spontaneous product-support conversations in named regional accents, including callers speaking over household noise. Assume an archive owner has office recordings containing the right product vocabulary, with permission to license a defined subset, but little of the required accent and noise combination. These are scenario assumptions, including the permissions, rather than observations about a provider's data.

Nora first writes down the model task and requested uses. She then asks for permission to inspect a sample and marks each requirement as present, missing or untested. The office material might be useful for domain language. It should not be priced as if it also solved the missing conditions. Cleaning the recordings or adding transcripts would improve their usability without creating the conversations that were never recorded.

The next decision is whether to license that useful subset and commission the gaps separately, or procure another collection that covers both. For a new collection, Nora can compare a managed project with contributor support in a workflow her team runs. She should keep a matching prebuilt dataset in contention too. The fictional owner can offer only the authorized subset, identify any export work and decide whether the requested exclusivity leaves enough value in its retained uses.

Bottom line: LXT's managed collection provides project coordination for new observations. That is the stronger fit when the missing material must be created and the buyer needs that coordination. Existing-data licensing fits when the eligible asset already contains the needed behavior.

Ask both: "Which sample will demonstrate our required conditions, which tests may we run, who accepts the result, and what happens to material that does not pass?" A volume total without those answers is too little to approve the purchase.

2. What will the buyer spend and the owner keep?

The buyer needs a complete cost for usable data. The owner needs a payment basis for the rights and work it supplies. A project budget, an owner's percentage and a contributor's task payment describe different transactions, even when each eventually produces a speech dataset.

LXT says most custom speech projects range from $15,000 for focused-domain datasets of 50 to 200 hours to $150,000 or more for multi-accent, multi-environment collections exceeding 1,000 hours. The page uses a dollar symbol without naming the currency. Confirm it before comparing quotes. This is a scoped project example with an open upper end, not an hourly tariff or a price for CaaS. Confirm monetary terms in the proposal.

For Rightmark's premium or bespoke route, the missing number is the denominator. Its homepage cost answer states the 75% to 85% owner share, but the owner still needs the applicable calculation base and any charges outside it written into the proposal. An attractive percentage is difficult to assess until you know what it multiplies and what event makes the proceeds payable.

Use the worksheet below for each proposal. Keep an included preparation charge inside the provider total; add it separately only when it is excluded. Treat internal review time as a real staffing requirement even if it never appears on the supplier invoice.

For the buyer, compare the quoted license or service plus separate charges, internal work and the agreed cost of changes. For the owner, compare proceeds alongside the preparation effort and rights committed. Keep those two summaries on separate lines; the owner's proceeds are one component of the transaction, not a measure of everything the buyer spends.

Bottom line: Public figures do not establish a price winner. Ask each provider to quote the same required output and rights, then account for the work left with your team. A higher proposal may include production you otherwise have to staff; a lower proposal may suit a team that already has that capacity.

Ask both: "For this deliverable and use, give us the full buyer cost, the owner or contributor payment basis, included work, payment milestones and the cost of failed acceptance or a scope change."

3. Which rights move, and what must stay out?

A buyer needs permission for the actual model work. An owner needs to know which records and future uses the agreement commits. A confidentiality agreement, contributor consent, ownership evidence and a license each answer a different part of that question.

An owner should work through the exclusion list before discussing the full archive's volume. Separate records it created from material supplied by customers, contractors or guests, then identify the authority for the proposed use of each subset. If a release covers one purpose, ask whether it covers the actual license being proposed. The unresolved records can remain outside the scope while the eligible material is assessed.

In an October 20, 2026 rights-screening exchange, Rightmark narrowed an expert-conversation archive to the subset with guest releases permitting third-party licensing; the remainder was held until updated releases were signed. The practical lesson is to update the eligible description when permissions change, so the buyer evaluates the same collection that the owner can authorize.

For buyers, write down what evaluation entails before copying a sample into a model workflow. Visual inspection, embeddings, benchmarking and fine-tuning are different requested activities. Rightmark's public buyer NDA limits portal material to deciding whether to license it; the exact non-code model tests need written permission. Use the sample-pack checklist to record the review question and boundaries.

Bottom line: Rightmark's Trust page gives owners concrete route distinctions to inspect. Its code-route own-use and indemnity provisions can matter more to an owner than the headline proceeds. For a new collection, obtain the resulting rights as carefully as the recording specification; technical fit alone is insufficient.

Ask both: "Which records and uses are licensed, what does the owner retain, and which exclusions, onward rights, warranties and liabilities appear in the agreement?"

4. Who can approve, change or stop the work?

Approval should name an event. Signing a standing license, approving a sample and authorizing more contributor tasks have different consequences. Owners and buyers should be able to point to who can say yes at each stage and what that decision permits next.

For an owner, write two separate instructions: what may be shared for evaluation and what may be licensed after signature. On Rightmark's code route, the signed agreement can authorize later non-exclusive sales. Someone who wants to approve every downstream sale should resolve that requirement before signing. Stopping new access is also a different request from ending an existing license; state what happens to each copy and each permission.

For a buyer choosing CaaS, decide who updates instructions when reviewers disagree. A task owner should be able to distinguish an unclear guideline from a genuine quality problem before commissioning more work. In a managed project, require an equivalent decision path with the project team: what changes, who approves it, and whether the price or delivery date changes too.

Bottom line: CaaS keeps task design and quality control with the client. That can suit a team with a mature process and people assigned to run it. Owners considering a standing license need equal clarity about the approvals they give once and the decisions they retain.

Ask both: "Who approves the source, sample, task changes, full-volume work and acceptance, and who can pause each stage without changing rights already granted?"

5. Can the source cover the languages and conditions we need?

A label such as audio or video is the beginning of a specification. The purchase needs to match the language, behavior, environment and history that matter to the model. An owner should describe the distinctive content it can demonstrate, then let the buyer assess whether that distinction serves its task.

Return to Nora's invented speech decision. She could make a coverage grid with accent groups as rows and recording environments as columns. Each cell would say what the sample actually demonstrates, what remains untested and what is absent. The office recordings still leave the noisy-home condition untested.

The grid also prevents a bad collection brief. Asking for more speech in a language leaves contributors guessing at the missing combination. Asking for a particular speaking condition gives the collection team something to design and the buyer something to inspect. Where a prebuilt collection fits that cell, compare it before commissioning the same observations again.

Bottom line: LXT can organize new collection to specified demographic and technical requirements. That is a meaningful advantage when the gaps require new observations. An existing archive is compelling when its eligible history or behavior is exactly what the task needs. Confirm the combination of conditions, not merely the category.

Ask both: "Show the coverage for our languages, accents, environments and behavior. What exists now, and which gaps require collection or another source?"

6. Who does the preparation and quality work?

The person paying for a task is not necessarily the person performing it. Buyers need to budget the work their own staff retain, and owners need to understand the exports and review they must support. Put names against the tasks before treating a proposal as fully managed.

In a February 27, 2024 Unite.AI interview, Carolyn Harvey, then LXT's COO, described "aligning with customers on success factors during the scoping phase". That is useful advice for either route: the definition of acceptable work belongs in the assignment, while there is still time to choose who can deliver it.

Add four columns to your task list: performer, approver, output and charge. For masking, specify what the performer removes and what evidence the approver sees. For annotation, name the guideline, the reviewer and how disagreements are resolved. For receipt, identify the person who checks the delivered files against the agreed specification. If a row has no performer, you have found work that must be staffed or added to the quote.

Rightmark's owner says deals are generally tailored. The exact non-code preparation map, permitted tests and acceptance event therefore remain matters for the applicable scope. Paying for extraction does not by itself tell the buyer who exports, masks, labels or checks the files.

Bottom line: LXT's managed annotation provides coordinated setup and execution. It deserves consideration when a buyer needs that production team. A buyer choosing contributor access must staff its own decisions, while an existing-asset licensing proposal needs an explicit preparation map alongside the rights.

Ask both: "For sourcing, rights evidence, export, masking, annotation, QA, acceptance and rework, name the performer, approver, output and charge, including our team's work."

7. What proof supports the choice for our project?

A useful reference should resemble the job you are buying. Ask how the work was accepted and which responsibilities stayed with the client. Owners should also inspect an agreement that matches the asset and rights they are considering.

LXT's multilingual voice case describes an early check that calls were reaching the client's platform. The client also filtered calls before sending them to LXT for transcription. Those details make the case useful to a buyer designing a handoff: collection can proceed while the receiving system has problems unless someone owns the receipt check. Request the same clarity for your own pipeline.

In LXT's hate-speech evaluation case, the research institute ran quality assurance and issued pass/fail decisions; LXT billed the client and paid contributors only for work that passed. That is a concrete example of the client retaining domain judgment. Ask for the acceptance and payment arrangement in your proposal rather than assuming this case's arrangement applies to another engagement.

Both are supplier-authored accounts. They offer inspectable operating detail, with more value for this decision than an unrelated logo. For Rightmark, the next useful evidence is an authorized example showing the requested data type, delivery and terms.

Bottom line: LXT's published cases give a procurement team specific workflows to examine. Use the matching case to ask better questions, then choose the route whose actual sample, rights and staffing fit your task. Keep an initial commitment bounded by the evidence you can obtain.

Ask both: "Which completed example matches this data type and operating model, who accepted it, and what can we inspect or test before expanding our commitment?"

• Appraise before sharing an asset. The free appraisal works from an owner's answers without files or system access, and suggests an indicative range, readiness and routes. Use it to decide what to investigate next, with the asset and rights still subject to review.
• Separate owner proceeds from the buyer fee. Standard licenses carry no upfront owner cost or payout deduction; the buyer pays Rightmark's fee on top. Premium and bespoke deals state a 75% to 85% owner share, agreed in writing. Confirm the calculation base and payment event.
• Bring ownership evidence and samples into the decision. The route-specific ownership pack comes from owner documents and reports for counsel's review, alongside agreed scrubbed samples under NDA. Define the permitted tests before running them.
• Keep ownership while defining the license. Owners retain the asset; code is non-exclusive by default and signing can cover later sales, while archive lump sums are usually exclusive for a term. Inspect the rights retained and any downstream uses before committing.
• Approve delivery access and inspect receipt. Owners authorize connections or exports; batches arrive graded with metadata and ownership evidence, and buyers scan again on receipt. Name the non-code preparation and acceptance responsibilities in the work order.

Choose LXT when you need new observations and want someone to coordinate their creation. Its managed collection includes consulting, setup, execution and dedicated project management. A missing accent or recording condition is a concrete reason to investigate that route, provided you can specify what a satisfactory collection would contain.

Choose CaaS when you already have working instructions, reviewers and a quality process. You keep task design and quality workflows while LXT manages contributors and infrastructure. Budget for those retained responsibilities. A team without reviewers should compare the managed service before committing to a workflow it cannot run.

Managed annotation is also available when the raw material is suitable but needs labels. LXT's prebuilt offering can be relevant when a named dataset meets the task; ask for its sample, rights and availability. Its speech service describes a rights-cleared supplier network, so an existing speech owner should ask about supplier licensing terms. The clickworker route pays for completed, client-approved tasks; that does not answer the archive owner's royalty question.

Choose Rightmark when an existing owner-held asset can satisfy the requirement and the remaining job is a workable license and delivery scope. The buyer should have a permitted sample that demonstrates relevant content, an eligible collection that covers the task, and written uses and responsibilities it can accept. Some pooled lot categories remain in preparation, so obtain the source and timing for the actual request.

For an owner, the fit is an asset it already holds and can authorize for licensing. Define exclusions early, consider whether an exclusive term would restrict other plans, and inspect any partner rights. A Rightmark code agreement can cover later non-exclusive sales without fresh approval, which makes the signed scope consequential.

Rightmark's public cost answer places extraction, scrubbing and transfer costs with the buyer. The deal still needs named performers, included work, acceptance and a payment event. Pursue the license when those answers make the asset usable for the buyer and worthwhile for the owner.

Consider a third option if

Your organization already has sufficient data, documented permission and people who can prepare it. Price the remaining internal work before buying another source. If an existing collection covers only part of the requirement, compare separate licensing and collection scopes with compatible permissions and one acceptance plan.

Hold or narrow the purchase if important records lack authority or the proposed use cannot be agreed. Rightmark does not currently take health data. A buyer with that requirement should assess a specialist against its actual data and handling needs.

Relevant reading: Rightmark vs Appen and off-the-shelf vs custom training data.

Buyers should choose the asset and operating model together. License suitable existing material, commission missing observations, or retain task design and QA while obtaining contributor support. A mixed scope can work if the rights, handoffs and acceptance decisions agree.

Owners should judge the license by the eligible asset, retained uses, preparation obligations and proceeds. Take the Ask both questions into the proposal discussion and record the answers in the worksheet. A narrower license or a held purchase is a useful result when the proposed scope does not fit.

Start with the data your model needs.

Tell Rightmark the data, intended use and volume you need. Rightmark replies with what it can source, volumes and timing, with sample timing confirmed during scoping.

Request data

Public pages below were checked November 3, 2026 unless noted. A displayed update date describes a revision, not the page's original publication date.

• Rightmark, homepage questions. Updated October 26, 2026.
• Rightmark, buyer process and ownership pack. Updated October 21, 2026.
• Rightmark, Trust and compliance. Updated October 18, 2026.
• Rightmark, free appraisal. Updated October 4, 2026.
• Rightmark, sample-pack checklist. Updated October 18, 2026.
• LXT, Data collection services. Published April 12, 2022; modified July 31, 2026.
• LXT, Crowd-as-a-Service announcement. Published July 16, 2026.
• LXT, Data annotation services. Published May 4, 2022; modified July 31, 2026.
• LXT, Speech recognition datasets. Published March 3, 2026; modified July 8, 2026.
• LXT, AI data solutions. Published December 23, 2025; modified October 19, 2026.
• LXT, collection quality statement. Published January 13, 2023.
• clickworker, contributor payment description. Published March 15, 2017; modified July 31, 2025.
• LXT, multilingual voice collection case. Published July 27, 2026.
• LXT, hate-speech evaluation case. Published July 24, 2026.
• Unite.AI, Carolyn Harvey, Chief Operations Officer at LXT: interview. Published February 27, 2024; checked November 2, 2026.
• Rightmark owner: direct answers, November 2, 2026. Used for the status of developing terms.
• Rightmark: anonymized rights-screening correspondence, October 20, 2026; permission-limited subset, without a transaction outcome.