TwelveLabs' Jockey went GA at $29 for 50 GB. The research preview gave 100 GB for $20.

That is how TwelveLabs billed Jockey on the Product Hunt board it won on 16 October, with 357 upvotes 1. The listing links to a page whose title reads "Jockey by TwelveLabs: AI Marketing Intelligence for Video" 2.

Jockey reached general availability on 15 October, after a research preview that opened in July 3. The search works. The meter that arrived with it is worth reading twice.

Jockey is three parts stacked. Marengo is the model that finds footage by meaning. Pegasus describes what happens inside a clip. Jockey runs a search loop across both and answers with timestamps 3.

You use it as a connector. Add Jockey to Claude or ChatGPT, point it at a library, and ask questions there 4. Most of the seven updates in the release concern that surface: ChatGPT support, imports from YouTube links and Google Drive, scene breakdowns drawn inside the conversation, sub-second clip boundaries, timeline editing, MP4 export, shareable artifacts 4.

The footage itself sits on TwelveLabs. You upload it once, and it lives in a knowledge store the platform indexes and keeps queryable 5.

The capability TwelveLabs leads with is entity recognition. Jockey finds the same person across thousands of videos and photos and groups everything that person appears in 6. That is the pitch: one search over an archive, rather than one search per file.

The problem is real. A phone camera roll and a marketing team's back catalogue are the same problem at different scales. Neither has ever been watchable in full.

Two of the three tiers now hold less and cost more. The free tier lost 40 percent of its storage. Plus lost half of its storage and gained $9 on the bill. Pro lost 300 GB and kept a dollar.

That comparison is clean for anyone putting a Jockey subscription against a Jockey subscription. It is less clean elsewhere, because the new plan also absorbs API usage that Developer-plan accounts used to be billed for separately. TwelveLabs says as much itself: "Your bill under the new billing can differ from your bill under legacy billing" 9.

Storage is published in gigabytes. The monthly Workload allowance is published as a word. The docs say the platform values your usage at the pricing page's rates and counts that total against the allowance 10. The pricing page names that allowance Basic, Medium or High and prints no figure for any of the three 7. You can work out how many gigabytes you are buying. You cannot work out how many answers.

The per-day response counts from the preview did survive, in a different document. Free still gets 15 Jockey responses a day, and Plus and Pro share 200 11. Overage is a third document again: available on Pro, off by default, at $0.065 per GB per month of storage, $2 per million Jockey tokens, $1.75 per hour of Pegasus analysis and $4 per thousand searches 1012.

New billing started on 15 October 2026. Free and Developer customers move on 4 January 2027. Developer customers pick Plus or Pro by 3 January, or land on Free by default. A monthly Jockey subscription moves on your first billing date in January 2027, and an annual one keeps its current terms until the term ends 9.

TwelveLabs is careful with one group. "TwelveLabs does not start a paid plan for you without your consent," the migration guide says, and that sentence belongs to Developer-plan users, who also get a deadline and a choice 9.

Jockey subscribers get no such sentence. Their required action in the same table reads "Nothing" 9. The guide also closes the early door: "You cannot move early while you have a paid Jockey subscription. This applies to monthly and annual subscriptions" 9. Cancelling is allowed, and a scheduled cancellation survives the move. Arriving on the new plan on a date of your own choosing is not.

An asset uploaded before your plan moves is what the docs call a pre-migration asset. It stops counting toward your storage allowance, stays accessible, and keeps its existing indexes 13.

The same section takes something back. A pre-migration asset keeps every Managed Intelligence benefit "except free re-indexing" 13. Free re-indexing is how a library keeps pace with new models. Every model release is sold as an automatic Jockey upgrade: "When Marengo improves, Jockey's search improves" 6.

Putting an old asset into a knowledge store, or into an index with Marengo turned on, requires a setting called storage accounting. Leave it off and the request fails with a 409 and the code asset_storage_accounting_required 13.

The setting runs one way. "Enabling storage accounting is permanent. You cannot turn it off. The asset counts toward your Managed Intelligence allowance from the moment of the call" 13.

Storage is measured as the average of your daily peak usage across the billing period 7. An archive that predates the change therefore sits outside the meter and outside the current models at the same time. The only bridge across is an irreversible claim on a smaller allowance than the one the archive was collected under.

Clause 4(a) of the Terms of Use takes a worldwide, royalty-free, sublicensable licence over everything you load into the services. The first purpose listed is to "provide, maintain, enhance, train, and improve" them, "including the training, re-training, fine-tuning, and improving the Twelve Labs Services unless you opt out of training through your account settings" 14.

Training is on until you find that setting and turn it off. For a product whose launch essay is about a father's camera roll, clause 4(a) is the one to read beside the essay 15.

Two more clauses sit in the same document. Clause 4(c) grants a licence over your name and logo for marketing, triggered by placing an order 14. Clause 1(b) reserves the right to change any element of the services at any time, without notice 14.

The privacy policy describes where the footage goes next. Customer data and derived output, including captions and vector representations, go to service providers and to "vendors that provide artificial intelligence services", in the United States, the Republic of Korea and elsewhere 8. Where TwelveLabs processes personal information on a customer's behalf, the customer is the controller, and the policy says TwelveLabs relies on customers to obtain consent 8.

Entity recognition is what makes that last sentence matter. Jockey groups one person across thousands of files 6, and the consent for everyone appearing in those files stays with whoever uploaded the library.

The launch essay opens on a father's phone and his children's first steps. Its problem statement is private: "the more we accumulate, the harder it becomes to find what matters." It ends on what software cannot recover: "There's no model or agent that can tell me what it felt like to hold my kids when they were little" 15.

The page that essay sits on sells to media buyers. Its title is "AI Marketing Intelligence for Video" 2, and the feature list is ad operations: Automatic Ad Tagging by hook, talent, format and tone, Winning Hook Detection, Talent and Logo Recognition, Format and Trend Breakdown, Performance Insights, and "Social & Trend Intelligence" drawn from "a social corpus we keep current" 2. TwelveLabs sells the same idea harder on its advertising page, which offers contextual ad breaks placed where attention pauses and reports a 10 to 20 percent lift in completion rates in side-by-side tests 16.

The announcement names both customers in a single sentence, running from "enterprise team to developers to creators" 3, and the release notes add social ingestion and per-video engagement metrics to what a library stores 34.

A parent's archive and an advertiser's library are one product here, counted in the same gigabyte. The plan built to carry both is the one that shrank at launch.

Jockey solves a real problem and puts the answer inside software people already pay for: it finds a moment in a library nobody has watched, and it returns timestamps you can check. Slow down on the plan. General availability handed the free tier 3 GB where the preview gave 5, and Plus $29 for 50 GB where the preview gave 100 for $20, while the monthly Workload allowance ships as a word whose number appears only after you buy. The library you already uploaded sits outside the meter and outside free re-indexing at the same time, and the one-way storage-accounting setting is the only bridge back to current models. Point Jockey at a working set you can afford to re-upload, keep the master archive somewhere else, and turn training off in your account settings before the first upload.