Jockey by TwelveLabs Review (2026): Pricing, Features & Honest Verdict

Jockey by TwelveLabs is an agentic video AI that searches your entire video and image library from a plain-English prompt and hands back the exact moments with frame-accurate timestamps. We found it to be the sharpest video-understanding layer we have tested. Best for marketing and creative teams buried in footage. Price: free up to 600 minutes, then pay-as-you-go from $0.042 per indexed minute. Rating: 7.5/10.

What is Jockey by TwelveLabs?

Jockey is the agent layer on top of TwelveLabs, a video-intelligence company founded in 2021 by Jae Lee and a team of researchers out of the machine-learning world. You point it at a library of videos, you ask a question in normal language, and it plans its own steps to find the answer. Not a folder. Not a filename. The actual second where the thing you asked about happens.

It started life as an open-source project on GitHub built with LangGraph and the TwelveLabs API. The version on twelvelabs.io/jockey today is the commercial evolution of that, currently in research preview with limited sign-ups. The timing is not an accident. TwelveLabs raised a $100M Series B in July 2026, co-led by NEA and Naver Ventures with Amazon and Index Ventures in the round. That is real money behind the video-search bet, and it shows in how polished the underlying models feel.

Here is the thing most video tools get wrong. They tag video with generic labels and call it search. Jockey reasons across the footage using multimodal models, so a query like “find the moment the founder mentions pricing objections” actually lands on that clip. That is a different category of tool.

What are Jockey’s key features?

Natural-language search across your whole library

This is the headline. You type what you are looking for and Jockey returns the specific clips, ranked, with timestamps you can deep-link to. It works on description, not metadata, so you do not need anyone to have tagged the footage first. For a team sitting on hundreds of hours of raw video, this alone changes the workflow.

Automatic ad and content tagging

Jockey can tag footage by hook, talent, format, and tone, and it recognizes specific people and logos across clips. If you are a performance-marketing team trying to work out which creative elements drive results, this turns a manual tagging job into a query. Ask it to pull every ad where a particular presenter appears in the first three seconds, and you get them.

Frame-accurate timestamps and structured extraction

Every result comes back with an exact timestamp, not a rough guess. You can also extract structured data against a JSON schema, so Jockey becomes a way to turn unstructured video into rows you can actually use downstream. That is the part developers will care about.

Multimodal embeddings and model integration

Under the hood it runs on TwelveLabs’ Marengo and Pegasus models and produces embeddings across video, image, audio, and text. It already integrates with Claude, and the team lists ChatGPT support as coming. So you can wire video understanding into an agent workflow rather than treating it as a standalone search box.

How much does Jockey cost?

Jockey itself is in research preview with no standalone price yet, but it runs on the TwelveLabs platform, and that pricing is public and refreshingly specific. The free plan gives you $0 cost, roughly 10 hours (600 minutes) of indexing, no credit card, 90-day data retention, and access to the search, embed, and analyze APIs within limits. That is a genuinely usable free tier, not a demo.

After that it is pay-as-you-go. Video indexing runs $0.042 per minute, search is $4 per 1,000 queries, and the analyze API is $0.0292 per minute of input video plus a small token charge on output. Enterprise pricing is custom. Compared to the big cloud vision APIs, which sit around $0.10 per minute for basic label detection, TwelveLabs is cheaper per minute and does far more than label the frames. You are paying for reasoning, not just detection.

Who is Jockey best for?

Use Jockey if you are a marketing or creative team with a large video archive and no fast way to find anything in it. The clip-discovery use case is where it earns its keep immediately. It is also a strong fit for AI-native creators who need to resurface old footage, and for developers building products that need to understand video, since the API and SDK are the real foundation here.

Skip Jockey if you only have a handful of videos. The overhead of indexing and learning the platform is not worth it for a small library you already know by heart. And if you need a fully generally-available, locked-down product with an SLA today, note that the hosted Jockey agent is still research-preview, so you may want to build directly on the stable TwelveLabs APIs instead.

What are the best Jockey alternatives?

Google Cloud Video Intelligence API. Usage-based at roughly $0.10 per minute for common label and shot detection, with a small free monthly tier. It is reliable and mature, but it labels video rather than reasoning across it, so you get detection without the agentic query layer.

Amazon Rekognition Video. Also usage-based, around $0.10 per minute analyzed, with a free tier for the first months. Good if you are already deep in AWS and want face and object detection, weaker if you want to ask open-ended questions of your footage.

Coactive AI. An enterprise multimodal video-intelligence platform with quote-based pricing. It targets large content operations and competes on the same “understand the whole library” promise, but there is no public price and it is aimed at bigger budgets.

Final verdict: Is Jockey worth it?

Jockey is one of the few video AI tools where the demo matches reality. The natural-language search actually finds the right moment, the timestamps are accurate, and the pricing is transparent in a market that loves to hide it. Backed by a company that just raised $100M, this is not a weekend project that will vanish. If your team wastes hours scrubbing through footage, the free 600 minutes will pay for itself in the first afternoon.

The caveats are honest ones. The hosted Jockey agent is still research preview, so access is gated and the product will keep shifting under you. There is also naming confusion between the open-source repo and the commercial page, which can trip up first-time users. Neither is a dealbreaker. For anyone whose real problem is finding things inside video, Jockey is worth getting on the waitlist for and building on today. Recommended, with the understanding that you are early.