Twelve Labs pricing

About

Twelve Labs is a video-understanding company that builds multimodal foundation models for search, retrieval, and analysis of video content. Its two flagship models are Marengo, a video foundation model for indexing, embeddings, and any-to-any retrieval (analyzing frames, temporal relationships, speech, and sound), and Pegasus, a video-first language model that fuses visual, audio, and speech signals for text generation and analysis. The models are delivered as a developer API with a hosted playground.

The company serves developers and enterprises building video search, content moderation, media-and-entertainment workflows, advertising, and government applications. Its pricing is positioned around the cost of processing video at scale rather than per-seat licensing, which aligns the bill with the actual quantity of video a customer indexes and analyzes.

Twelve Labs is privately held and venture-backed. It raised a $5M seed led by Index Ventures in 2022, a $12M seed extension alongside an Oracle partnership, a $10M strategic round in October 2023 from Nvidia, Intel, and Samsung Next (bringing total funding to roughly $27M), and a $50M Series A in June 2024 co-led by NEA and NVIDIA’s NVentures, with Intel Capital, Index, Radical Ventures, and Korea Investment Partners participating. By late 2023 the company reported around 17,000 developers and named customers including the NFL across media, entertainment, advertising, and government.

It positions against general-purpose multimodal LLM APIs by specializing in long-form video understanding, and competes for the same developer budget as cloud video-intelligence services (Google Cloud Video Intelligence, AWS Rekognition Video, Azure Video Indexer). Its models are also distributed through Amazon Bedrock, announced in 2025.

Pricing summary : How Twelve Labs meters video by the minute

Twelve Labs uses a pure usage (pay-as-you-go) model on its Developer plan, metered primarily by the minute of source video, with a free entry tier and custom committed-use enterprise contracts. The core dimensions:

1. Marengo indexing & infrastructure: $0.042 per minute ($2.50/hour) for video indexing, plus $0.0015 per minute ($0.09/hour) for embedding infrastructure, billed monthly.
2. Search API: $4 per 1,000 queries.
3. Embed API (one-time, by input type): Video $0.042/min, Audio $0.0083/min ($0.50/hour), Image $0.1 per 1,000 requests, Text $0.07 per 1,000 requests.
4. Pegasus analysis: $0.0292 per minute of input video plus $0.0075 per 1,000 output text tokens.

What makes this different: the headline meter is the duration of video rather than tokens or API calls, and the pricing page lets buyers flip every rate between a per-minute and a per-hour display of the identical underlying price. It sits among the pure-usage AI platforms in our corpus and is packaged like other developer-tools pricing — public self-serve rates with a gated enterprise tier on top.

Pricing by product

Plans (Free / Developer / Enterprise)

Marengo (indexing, infrastructure, search) — Developer per-unit rates

Embed API (one-time, by input file type) — Developer per-unit rates

Pegasus (analysis) — Developer per-unit rates

Sales motions across products: PLG / self-serve for the Free and Developer (pay-as-you-go) plans; sales-led for Enterprise committed-use contracts and model fine-tuning.

Hidden costs : What heavy video-indexing workloads actually pay

The Developer plan stacks several separate meters — one-time indexing, a recurring monthly infrastructure fee, search, the Embed API per input type, and Pegasus analysis plus output tokens — so a single workflow can touch four or more line items at once. Two things surprise first-time buyers: the infrastructure fee recurs every month for as long as embeddings are stored (it is not a one-time charge like indexing), and Segment calls multiply the billed duration by the number of segment definitions in the request.

Archetype 1 — a media team indexing 600 hours/month and running light search. Using the published Developer rates (mirrors the official calculator at 600 hrs = 36,000 min indexing, 1,000 search queries, 1,000 min Pegasus analysis):

The indexing charge dwarfs everything else, but note the $54 infrastructure line recurs every month even if no new video is indexed — storage of embeddings is a standing cost, not a sunk one.

Archetype 2 — a Segment-heavy analytics workload. A team that runs Segment with 4 segment definitions on each clip, and frequently omits the time-range parameters, pays the full video duration × 4 per call. A nominally “15-minute” review of a 60-minute asset that omits start/end bills 240 minutes (60 × 4), not 15 — a 16× multiplier on the headline rate before a single token of output is generated.

Want to estimate your own Twelve Labs bill? Use the Twelve Labs pricing calculator to model your monthly cost based on indexed video minutes, search queries, and Pegasus analysis volume — and cross-check it against our guidance on tracking and metering usage events.

Pricing evolution : From contact-sales to a per-capability grid to one flat per-minute rate

Twelve Labs moved from a contact-sales-only page in 2022 to public self-serve rates in 2024, then through a major repackaging in 2025 that collapsed a per-capability indexing grid into a single flat per-minute rate. Prices below were read from archived snapshots of the pricing page.

Cadence

Tracked range: 2022-Q2–2026-Q2. Quarters not listed were verified stable (0 price changes, 0 SKU additions).

Notable changes

• 2022-05 — Pricing page is Contact-sales only with no published per-minute rates; a banner announces the $5M seed led by Index Ventures.
• 2023-01 — Banner announces a $12M seed extension and an Oracle partnership; the page is still Contact-sales.
• 2024-04 — First public self-serve rates: per-capability one-time indexing (Visual $0.033/min, Conversation $0.0083/min, Text-in-video $0.067/min, Logo $0.10/min), Pegasus output text $0.002/1K tokens, monthly infra $0.0015/min, Free index limit 600 min.
• 2025-05 — Adds an Embed API priced per embedding (Video $0.003/embedding) and daily API call caps (Search 3,000/day on Developer); indexing rates unchanged.
• 2025-08 — Major repackaging to flat per-minute metering ($0.042/min indexing), search re-priced to $4/1,000 queries, Embed moved to per-minute, Pegasus Analyze input video at $0.021/min; banner announces models in Amazon Bedrock.
• 2026 — Pegasus Analyze input video raised to $0.0292/min (legacy Pegasus 1.2 still billed at $0.021/min per the FAQ).

The 2025 repackaging in detail

The single largest pricing change in Twelve Labs’ history was the mid-2025 shift from per-capability metering to a flat per-minute rate. Through 2024 and into mid-2025, indexing was billed by what the model extracted: Visual at $0.033/min, Conversation/Audio at $0.0083/min, Text-in-video at $0.067/min, and Logo at $0.10/min — so the cost of indexing one minute of video depended on which capabilities you turned on, ranging from $0.0083 to over $0.20/min if you enabled everything. By the 2025-08 snapshot, all of that had collapsed into a single $0.042/min indexing rate regardless of capability. Search was simultaneously re-based from daily call caps to a clean $4 per 1,000 queries, and the Embed API moved from per-embedding to per-minute pricing. The repackaging coincided with the Amazon Bedrock launch, where a single uniform rate is far easier to list in a marketplace catalog than a four-way capability matrix. The one later adjustment — Pegasus Analyze rising from $0.021 to $0.0292/min — is the only headline rate that has gone up since the flat model launched.

What’s unique : Per-minute video metering and a dual-model split

1. The meter is the minute of video. Almost every AI-API peer in the pricing blueprint meters tokens, requests, or seats; Twelve Labs meters the duration of source video — rounded up to the next full minute of successfully indexed footage. That makes the bill legible to a media buyer who thinks in hours of content, not tokens, and it ties cost directly to the unit of value being processed. It is one of the cleanest examples in the corpus of choosing a value metric that matches how the customer already counts their own work.

2. Per-minute / per-hour display toggle. The pricing page lets a buyer flip every rate between a “Pricing by minute” and “Pricing by hour” view of the identical underlying price — $0.042/min renders as $2.50/hour. This is a pure presentation affordance with no pricing consequence, but it meets two different mental models (developers think in minutes, content owners think in hours) without forcing either to do the arithmetic.

3. Marengo vs Pegasus split billing. The two models are billed as separate product lines with separate meters: Marengo covers indexing, embeddings, and search; Pegasus covers analysis (input video) plus generated output tokens. A workflow that both searches and summarizes touches both meters at once, which improves cost attribution but means no single headline number captures the true cost of an end-to-end task. This is the trade-off at the heart of every multi-meter usage-based pricing model: alignment versus forecastability.

Strengths & weaknesses

Billing UX : Self-serve metering controls and the cost estimator

• Pricing calculator — a dedicated /pricing-calculator surface estimates monthly cost from inputs for indexed video (hours/minutes), monthly search queries, Embed API usage (video, audio, image, text), and Pegasus input-video duration plus output tokens, with a 1–12 month projection multiplier.
• Per-minute / per-hour display toggle — the pricing page switches every rate between a “Pricing by minute” and “Pricing by hour” view of the same underlying price.
• Compare plans grid — surfaces capacity limits per plan: video hours usage, index access duration, duration per index, videos per index, and concurrent indexing tasks.
• Self-serve upgrade — registering a credit card in the Dashboard upgrades Free to the pay-as-you-go Developer plan; mid-month infrastructure fees are prorated from the upgrade date.
• Dashboard billing controls — the plan downgrade / cancel-enrollment flow lives under Dashboard → Billing, with documented behavior for index retention on downgrade.

Strategic wins : Decisions that aligned price with video cost

1. Metering by the minute of video

Choosing the minute of source video as the primary meter aligned the bill with the exact unit a media or content buyer already counts. Unlike token metering — which is opaque to anyone who does not think in model internals — a per-minute rate is something a customer can sanity-check against their own content library. It is a textbook case of picking the value metric the customer already understands, and it let Twelve Labs publish a calculator that a non-engineer can use.

2. A free, non-expiring 600-minute allowance

The Free plan grants a one-time, accumulated 600 minutes (10 hours) that does not decrease when you delete videos — so an evaluator can experiment, delete, and re-try without burning the allowance on mistakes. Requiring no credit card removes the highest-friction step in developer onboarding. This is a strong product-led on-ramp: the allowance is generous enough to ship a prototype, but the 90-day index-access expiry on Free creates a natural, non-punitive nudge toward the Developer plan once a project goes real.

3. Simplifying to one flat per-minute rate

The 2025 collapse of the four-way capability grid (Visual / Conversation / Text-in-video / Logo) into a single $0.042/min indexing rate traded some price precision for enormous legibility. Buyers no longer had to model which capabilities they would enable to estimate a bill, and the uniform rate made the models trivially listable in the Amazon Bedrock marketplace. The lesson for other usage-priced teams: a simpler packaging that a buyer can compute in their head often beats a granular one they cannot.

Areas to improve : Forecastability and tier clarity

1. Reduce the number of separate meters a single workload touches

An end-to-end search-and-summarize task can hit indexing, monthly infrastructure, search, and Pegasus analyze plus output tokens — four to five line items for one user-visible action. A “bundled workflow” rate (e.g., a single per-minute price for “index + analyze”) would let buyers reason about the cost of a job rather than reverse-engineering it from a stack of meters. The forecastability problem here is the same one we describe for unpredictable AI product costs.

2. Make the recurring infrastructure fee unmissable

The monthly $0.0015/min embedding-infrastructure fee recurs for as long as embeddings are stored, but it sits visually beside one-time indexing rates on the page. Surfacing it as an explicit “standing monthly cost” line — and showing it prominently in the calculator’s summary — would prevent the common surprise of a bill that does not drop to zero in a month with no new indexing.

3. Publish indexed-minute bundles or commitments for predictability

Today the only commitment path is a fully gated Enterprise contract. A published middle tier — prepaid minute bundles or a committed-use discount visible on the page — would give mid-market buyers a way to trade volume for a lower rate without entering a sales cycle, the same pattern we recommend in understanding prepaid credit models.

Key takeaways

1. Meter on the unit of value your customer already counts. Twelve Labs bills the minute of video, the same unit a media buyer uses to describe their own library. A meter the customer can verify against their own data is worth more than a technically precise one they cannot.
2. Simpler packaging can beat granular packaging. Collapsing a four-way capability grid into one flat per-minute rate cost some precision but made the bill computable in a buyer’s head — and listable in a marketplace. Granularity that no one can forecast is a liability, not a feature.
3. Free allowances should be forgiving, not just large. The non-decrementing 600-minute allowance lets evaluators delete and retry without penalty, which matters more for conversion than a bigger one-shot quota would.
4. Split models means split meters — budget for the attribution cost. Marengo and Pegasus bill separately, which is great for cost attribution but means no single number describes an end-to-end task. Expect to build a calculator to make a multi-meter model legible.
5. Publish your rates, gate only the negotiation. Twelve Labs publishes every self-serve rate and gates only Enterprise commitments — a transparency posture that lets developers self-qualify before ever talking to sales.

UBP implications

1. Duration is an underused value metric for media-AI. Where tokens dominate text and image APIs, duration (minutes of video/audio) maps cleanly onto how content businesses already account for their assets — a metric worth considering for any AI product that processes time-based media.
2. Multi-meter usage models trade alignment for forecastability. Splitting the bill across indexing, infrastructure, search, and analysis aligns each charge to a distinct cost driver, but the buyer pays in predictability. Teams adopting this pattern must invest in calculators and bundled-rate options to keep the model usable.
3. Repackaging is a pricing lever, not just a cleanup. Twelve Labs’ move from a capability grid to a flat rate was timed to a marketplace launch — a reminder that simplifying a usage model can unlock distribution channels that a complex one cannot support.

Sources

• Twelve Labs pricing page (accessed 2026-06-02)
• Twelve Labs pricing calculator (accessed 2026-06-02)
• Twelve Labs API docs (accessed 2026-06-02)
• Twelve Labs changelog (accessed 2026-06-02)
• Twelve Labs blog (accessed 2026-06-02)
• Archived pricing pages via the Wayback Machine: snapshots dated 2022-05-17, 2023-01-20, 2024-04-19, 2025-05-13, and 2025-08-10 (accessed 2026-06-02)

Browse the full pricing blueprint corpus for comparable usage-metered AI platforms.

Bottom line

Twelve Labs prices video understanding the way it is consumed — by the minute of source video — splitting the bill across Marengo indexing, infrastructure, and search and Pegasus analysis and output tokens, with a free 600-minute on-ramp and custom enterprise commits above it.

Want to compare Twelve Labs against other usage-metered AI pricing? Browse the pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Current pay-as-you-go: Pegasus analyze raised to $0.0292/min

Jun 2026

Live page meters Marengo indexing at $0.042/min, infrastructure at $0.0015/min, search at $4/1K queries, Embed API per input type, and Pegasus analysis at $0.0292/min (up from $0.021/min in 2025-08) + $0.0075/1K output tokens; free tier offers 600 one-time minutes; the FAQ notes legacy Pegasus 1.2 still bills analyze at $0.021/min.

Repackaged to flat per-minute metering; Amazon Bedrock availability

Aug 2025

Wayback 2025-08-10 shows the per-modality grid collapsed into a flat $0.042/min video indexing rate, search re-priced to $4/1,000 queries, Embed API moved to per-minute (Video $0.042/min, Audio $0.0083/min), and Pegasus Analyze input video at $0.021/min + output $0.0075/1K tokens. A banner announced TwelveLabs models are now in Amazon Bedrock.

Per-engine modality grid with added Embed API and daily call caps

May 2025

Wayback 2025-05-13 shows the same per-engine indexing rates (Visual $0.033/min, Audio $0.0083/min, Text-in-video $0.067/min, Logo $0.10/min) plus a new Embed API priced per embedding (Video $0.003/embedding, Image/Text $0.05 & $0.01 per 1K) and daily API call limits (Search 3,000/day, Embed 1K/day).

First public self-serve rates — per-capability indexing (Marengo-2.6 / Pegasus-1)

Apr 2024

Wayback 2024-04-19 shows a Free / Developer / Enterprise grid with per-capability one-time indexing: Visual $0.033/min, Conversation $0.0083/min, Text-in-video $0.067/min, Logo $0.10/min; Pegasus Generate output text $0.002/1K tokens; infra $0.0015/min; Free index limit 600 min, Developer 60,000 min soft limit.

Still contact-sales; $12M seed extension + Oracle partnership

Jan 2023

Wayback 2023-01-20 still showed no published rates behind a Contact sales CTA. The top banner announced a $12M seed extension and an Oracle partnership to bring the video-understanding model to market.

Contact-sales pay-as-you-go (no published rates) — $5M seed era

May 2022

Launch-era pricing page (Wayback 2022-05-17) showed only Flexible / Pay-as-you-go / Transparent positioning behind a Contact sales CTA — no per-minute rates published. A banner announced the $5M seed led by Index Ventures.