Comet pricing

About

Comet (legal entity Comet ML, Inc.) is an AI/ML developer-tools company that sells two complementary observability products under one roof. Comet MLOps is the original product: an experiment-tracking, dataset-versioning, and model-registry platform for teams building and training machine-learning models. Opik is the newer, LLM- and agent-era product: an open-source-first observability and evaluation platform for tracing, testing, and monitoring applications that make LLM calls. Both products run on the same underlying platform, and Comet markets them as one company with “two flagship product families.”

Comet serves a spectrum from individual researchers and students (it offers a free academic Pro plan) through growing engineering teams up to large enterprises that need SSO, dedicated support, and SOC 2 / ISO 27001 / ISO 9001 / HIPAA / GDPR compliance. Opik’s true open-source core (the same codebase as the hosted version, self-hostable from GitHub) gives it a developer-led adoption motion that sits alongside the more traditional self-serve and sales-led motions on the MLOps side.

The defining characteristic of Comet’s pricing is that the two products meter on entirely different value units — Opik on spans (LLM/agent trace units) and Comet MLOps on training hours and data storage — yet they are presented side by side on a single /site/pricing/ page via a product toggle. This makes Comet a useful corpus example of a multi-product company that resisted collapsing two genuinely different cost drivers into one generic “credits” abstraction.

Pricing summary : two products, two value metrics, freemium-to-seat

Comet runs a freemium model that steps into a seat fee, with usage metering layered on the MLOps side. The two product lines price independently across three to four tiers each, and the value metric differs by product — so it is best read as two pricing systems on one page rather than a single unified model. See how this compares to other hybrid pricing models in the corpus.

1. Opik (LLM/agent observability) — metered by spans: Open Source self-hosted ($0), Free Cloud ($0, up to 10 team members, 25k spans/month, 60-day retention), Pro Cloud ($19/month flat for the whole account, up to 50 team members, 100k spans/month, customizable span limits and retention), and Enterprise (custom, unlimited members).
2. Comet MLOps (experiment tracking) — metered by training hours + storage, per seat: Free ($0, 1 user, fair-usage limits, 100GB storage), Pro ($19 per user/month, up to 10 users, 1,500 training hours + 500GB included, then $1 per training hour and $3 per 100GB/month overage), and Enterprise (custom, unlimited users and training hours).

What makes this different: Comet charges the same $19 headline on both products but on two different denominators — flat-per-account for Opik versus per-user for MLOps — and keeps the LLM-observability and classical-ML value metrics deliberately distinct rather than forcing both into a single credit-based unit.

Pricing by product

Opik — LLM/agent observability (metered by spans)

Opik’s value metric is the span — a unit of an LLM/agent trace. Free Cloud includes 25k spans/month; Pro Cloud raises this to 100k spans/month and unlocks customizable monthly span limits and customizable data-retention periods. Pro Cloud and Enterprise add LLM-observability extras (online evaluation, AI guardrails, alerts); guardrails are available for self-hosted deployment.

Comet MLOps — experiment tracking (metered by training hours + storage)

Sales motions across products: PLG / self-serve for Opik (Open Source, Free Cloud, Pro Cloud) and Comet MLOps (Free, Pro); sales-led for both Enterprise tiers.

Comet MLOps usage overage rates

Beyond the Pro plan’s included allowances, Comet MLOps meters two usage dimensions explicitly (from the plan comparison table):

A training hour is 60 minutes of active model-training time recorded in Comet, measured between starting and stopping an Experiment. Opik plans, by contrast, expose no per-unit overage rate publicly — Pro Cloud simply raises the included span ceiling and makes the monthly span limit customizable.

Hidden costs : training-hour and span overage on Comet bills

The advertised $19 headline understates what a heavy Comet MLOps team actually pays, because training hours and storage meter beyond the Pro plan’s included allowances. One worked example using the captured overage rates ($1/training hour, $3/100GB/month):

Comet MLOps — 8-person team training heavily

At this volume the training-hour overage dwarfs the seat cost — the usage dimension, not the seat fee, is what scales the Comet MLOps bill. (Opik bills differently: it raises the included span ceiling on Pro Cloud rather than publishing a per-span overage rate, so heavy Opik usage tends to push teams toward an Enterprise custom plan.)

Want to estimate your own Comet bill? Use the Comet pricing calculator to model your monthly cost based on seats, training hours, and storage.

Pricing evolution : from per-seat MLOps to a dual-metric, two-product page

Comet’s pricing has moved through three eras. It began as a single MLOps product priced per seat on storage credits (Teams $179–$249 per user/month in 2022), then in late 2022 rebuilt MLOps around a training-hour meter (Starter $50/user/month). In September 2024 it added an entirely separate product — Opik, metered on spans — and through 2024–2025 it steadily cut the MLOps seat price (from $50 to $39 to today’s $19) while holding the usage allowances and overage rates constant. The result is the current two-product, two-value-metric page. The dollar figures below come from dated Wayback snapshots of /site/pricing/; where the JS-gated plan cards did not render a numeral, the cell is marked unknown and never guessed.

Cadence

Tracked range: 2022-07 – 2026-06 (MLOps page); Opik page tracked 2024-09 onward. “unknown” marks a quarter where the archived plan card preserved structure but the JS-gated dollar numeral did not render and was not independently legible. Quarters not listed were verified stable (0 changes, 0 additions).

Notable changes

• 2021-04-08 — $13M Series A led by Scale Venture Partners (Trilogy, Two Sigma participating). (Comet press release.)
• 2021-11-18 — $50M Series B led by OpenView. (Comet press release.)
• 2022-07 — Wayback shows MLOps priced per seat on storage credits: Teams $179/user/mo (250GB) and Teams Pro $249/user/mo (500GB), with a free Community tier.
• 2022-11 — MLOps repriced onto a training-hour meter: Starter at $50/user/month bundling 5,000 annual organization training hours, then $1 per training hour, with 500GB storage. This is the meter that still defines MLOps today.
• 2024-09-17 — Comet launched Opik, an open-source LLM evaluation and observability platform, adding a second product metered on spans. (Comet press release.)
• 2024-11 — MLOps moved from 5,000 annual to 1,500 monthly training hours, set Pro at $39/user/month, and introduced the explicit $3/100GB storage overage alongside $1/training hour.
• 2025-12 — MLOps Pro seat price cut from $39 to $19/user/month with allowances unchanged, lining the headline up with Opik Pro Cloud’s $19.

What’s unique : two value metrics under one pricing page

1. Two distinct value metrics on a single pricing page. Most multi-product companies collapse everything into one abstract unit (generic “credits”). Comet does the opposite: Opik meters spans (LLM/agent trace units) and Comet MLOps meters training hours plus storage, presented side by side behind one Opik/MLOps toggle. Each unit tracks what actually drives cost for that workload, which is the textbook resolution of the value-metric problem in AI pricing — pick the metric the buyer can reason about rather than the one that maximizes per-action revenue.

2. The same $19 headline on two different denominators. Opik Pro Cloud is $19 flat per account (up to 50 team members), while Comet MLOps Pro is $19 per user per month. The number is identical; the denominator is not — a flat-per-account ceiling for the LLM-observability product versus a per-seat fee for the experiment-tracking product. It is a rare case of a vendor deliberately using one price point to mean two different things across two products on the same page.

3. True open-source Opik on the same codebase as the hosted product. Opik’s Open Source tier is not a stripped “community edition” — it is the same codebase as the hosted version, self-hostable from GitHub at $0. That gives Comet a developer-led adoption wedge in front of its traditional MLOps sales motion, the open-core variant of usage-based pricing in SaaS and AI where free usage seeds the funnel rather than cannibalizing it.

4. A usage meter that survived two repricings unchanged. Comet has cut the MLOps seat price twice (from $50 to $39 to $19) since 2022, yet the $1/training hour overage rate has held constant the entire time, and the $3/100GB storage rate has held since it appeared in 2024. The company moves the seat fee to stay competitive but keeps the usage dimension stable — a clean separation of the “access” price from the “consumption” price that makes the usage-based model legible across years of repricing.

5. It resisted a single generic “credits” abstraction. When Comet added Opik in 2024, the obvious shortcut would have been to fold spans, training hours, and storage into one “Comet credit”. It did not — and that restraint is what keeps each bill readable, because a training run and an LLM trace are genuinely different cost drivers that a unified credit would have blurred.

Strengths & weaknesses

Billing UX : span limits, training-hour metering, and the product toggle

• Opik | MLOps product toggle — the /site/pricing/ page presents both product families behind a single tab control; the buyer picks Opik or Comet MLOps and sees that product’s tiers, since the two meter on different value units.
• Customizable monthly span limits (Opik Pro Cloud) — Pro Cloud accounts can raise their monthly span ceiling above the 100k baseline rather than being hard-capped, giving teams a knob for scaling LLM-trace volume.
• Customizable data-retention periods (Opik Pro Cloud) — Pro Cloud lets accounts adjust the 60-day default retention window, a billing-relevant control because retained span data drives storage.
• Training-hour metering (Comet MLOps) — MLOps measures usage in training hours (60 minutes of active Experiment training time each), with 1,500 included on Pro and $1 per hour beyond, so the bill tracks actual model-training compute.
• Service accounts and view-only users (Enterprise) — Enterprise tiers add machine-to-machine service accounts and read-only seats, letting orgs separate automated logging and view-only access from billable interactive users.
• Compare Plans + Start Free Trial / Contact Us CTAs — the page exposes a full feature-comparison table and routes self-serve tiers to “Start Free Trial” while Enterprise routes to “Contact Us” / “Get Demo”, making the self-serve-vs-sales boundary explicit.

Strategic wins : open-source Opik and transparent overage rates

1. Matching each product to its true value metric

Comet resisted the temptation to unify two very different products under one generic “credits” unit. Opik meters spans (LLM/agent traces) and Comet MLOps meters training hours and storage — each unit tracks what actually drives cost for that workload, which keeps bills legible. This is a textbook example of choosing the right usage metric for the work being done, and a clean counterpoint to the value-metric problem in AI pricing.

2. True open-source Opik as a developer-led on-ramp

Opik’s Open Source tier is the same codebase as the hosted version, self-hostable from GitHub at $0. That gives Comet a credible developer-adoption motion that seeds later conversion to Free Cloud, Pro Cloud, and Enterprise — a product-led wedge on top of the traditional MLOps sales motion, similar to how other usage-based pricing in SaaS and AI plays seed free usage before monetizing.

3. Publishing explicit overage rates on the usage dimension

Comet MLOps states its overage rates in the open — $1 per training hour and $3 per 100GB/month — rather than hiding them behind “contact sales”. Transparent per-unit usage-based pricing lets teams forecast their own bills and builds trust before a quote.

Areas to improve : span overage opacity and dual-$19 confusion

1. Publish an Opik per-span overage rate

Comet MLOps exposes clean per-unit overage rates, but Opik does not publish a per-span price — Pro Cloud simply raises the included ceiling and makes the limit “customizable”. A published per-span (or per-1k-span) rate would let Opik teams model overage the way MLOps teams already can.

2. Disambiguate the two $19 plans

Opik Pro Cloud ($19 flat per account) and Comet MLOps Pro ($19 per user/month) share a headline number but bill on different denominators, which is easy to misread on a single toggled page. A clearer per-plan label (“$19/account” vs “$19/user”) would reduce buyer confusion at the comparison step.

3. Make the pricing page render without the JS tab gate

The /site/pricing/ plan tables are gated behind a JS-hydrated tab block that does not paint for non-interactive clients (and broke automated screenshot capture entirely). Server-rendering the plan cards would improve accessibility and make the prices indexable by AI search engines that read static HTML — a real concern given how unpredictable AI product costs already make buyers price-sensitive.

Key takeaways

1. Match the meter to the workload, not to the org chart. Comet runs two products on two value metrics — spans for Opik, training hours plus storage for MLOps — because a training run and an LLM trace are genuinely different cost drivers. Forcing both into one generic credit would have made bills harder to reason about, not easier.
2. A shared headline price can carry two meanings. Both Pro tiers are “$19,” but Opik’s is flat per account and MLOps’s is per user. Reusing a price point across products is a legitimate move, but only if the denominator is labeled clearly enough that buyers don’t misread it.
3. Separate the access price from the consumption price. Comet cut the MLOps seat fee from $50 to $39 to $19 while holding the $1/training-hour and $3/100GB rates steady. Treating the seat fee and the usage rate as independent levers let it stay price-competitive without re-litigating the whole model.
4. Open source can be the on-ramp, not the leak. Opik’s true-OSS tier (same codebase as cloud) gives Comet a developer-led wedge that feeds Free Cloud, Pro Cloud, and Enterprise — monetizing hosting, scale, and compliance rather than core capability.
5. Transparency is uneven, and that’s the gap. MLOps publishes its overage rates in the open; Opik does not publish a per-span rate. The half that’s transparent builds trust; the half that isn’t sends heavy users straight to a sales conversation.

UBP implications

1. Multiple value metrics beat one forced abstraction when cost drivers genuinely differ. Comet shows that a multi-product company can keep bills legible by metering each product on its real cost driver instead of collapsing everything into a single “credit.” The cost is buyer-education overhead; the benefit is that every line item maps to something the customer recognizes.
2. Decoupling seat price from usage price gives pricing teams room to maneuver. Because Comet’s overage rates are independent of its seat fee, it can re-price access (the seat) for competitiveness without touching the consumption economics. Usage-based packages that keep these two levers separate can adapt faster than ones that bundle them.
3. Publish the overage rate or accept that heavy users will route to sales. MLOps’s explicit $1/training-hour and $3/100GB rates let teams self-forecast; Opik’s missing per-span rate forces scaling teams into Enterprise. Whether to expose the marginal rate is a deliberate trade between self-serve forecastability and sales-led capture.

Sources

• Comet pricing page (Opik + MLOps) (accessed 2026-06-02)
• Comet Opik product page (accessed 2026-06-02)
• Comet documentation (accessed 2026-06-02)
• Comet blog (accessed 2026-06-02)
• Comet launches Opik — press release (2024-09-17) (accessed 2026-06-02)
• Comet $50M Series B press release (accessed 2026-06-02)
• Comet $13M Series A press release (accessed 2026-06-02)
• Opik on GitHub (accessed 2026-06-02)
• See how Comet compares across the pricing blueprint corpus (accessed 2026-06-02)

Bottom line

Comet is one of the cleanest corpus examples of a multi-product company refusing to flatten two genuinely different cost drivers into one meter: Opik prices LLM/agent observability on spans, Comet MLOps prices experiment tracking on training hours plus storage, and both sit on a single page behind one toggle. The same $19 headline means “flat per account” for Opik and “per user” for MLOps — a deliberate, if easily misread, reuse of one price point. Its MLOps overage rates ($1/training hour, $3/100GB) have outlasted two seat-price cuts, while Opik’s missing per-span rate is the one transparency gap that still pushes heavy users toward a sales call.

Want to compare Comet against other ML/LLM observability 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 snapshot — dual-product pricing (Opik + MLOps)

Jun 2026

Opik: Open Source $0 / Free Cloud $0 (25k spans) / Pro Cloud $19/mo flat (100k spans) / Enterprise custom. Comet MLOps: Free $0 / Pro $19 per user/mo (1,500 training hours + 500GB, $1/training-hour and $3/100GB overage) / Enterprise custom.

MLOps Pro cut from $39 to $19 per user/month

Dec 2025

Wayback snapshots show MLOps Pro at $39 per user/month through 2025-09, then $19 per user/month by 2025-12 — a ~50% seat-price cut while the 1,500 training-hour + 500GB allowances and $1/$3 overage rates held steady. This aligned MLOps Pro's headline with Opik Pro Cloud's $19.

MLOps moves to monthly training-hour buckets — Pro $39/user/mo

Nov 2024

Wayback snapshot (2024-11) shows MLOps restructured to Community (free), Pro ($39 per user/month, 1,500 training hours and 500GB included, then $1/training hour and $3/100GB overage), and Enterprise. The allowance shifted from 5,000 annual org hours to 1,500 monthly training hours, and a $3/100GB storage overage rate appeared.

Opik launches — open-source LLM evaluation & observability

Sep 2024

Comet launched Opik, an open-source, end-to-end LLM evaluation and observability platform (tracing, automated evals, pre-deployment testing) — adding a second product on a second value metric (spans) alongside classic MLOps. Opik is fully open source on GitHub with a hosted cloud and enterprise option. (Comet press release, 2024-09-17.)

MLOps repriced onto training hours — Starter $50/user/mo

Nov 2022

Wayback snapshot (2022-11) shows a repackaged MLOps: Community (free forever, 100GB), Starter / Most Popular ($50 per user/month, 5,000 annual organization training hours, $1 per training hour beyond, 500GB), and Enterprise (unlimited training hours). This introduced the training-hour meter that still defines MLOps today.

MLOps priced per-seat on storage credits — Teams $179 / Teams Pro $249

Jul 2022

Wayback snapshot (2022-07) shows Comet MLOps as Community (free, 1 member, 100GB storage credits), Teams ($179 per user/month, 250GB), and Teams Pro ($249 per user/month, 500GB), plus Enterprise. The value metric at this point was storage credits and seats — no training-hour meter yet.

$50M Series B (OpenView)

Nov 2021

Comet raised a $50M Series B led by OpenView (with Scale, Trilogy, and Two Sigma) to accelerate enterprise ML development. (Comet press release.)

$13M Series A (Scale Venture Partners)

Apr 2021

Comet raised a $13M Series A led by Scale Venture Partners, with Trilogy Equity Partners and Two Sigma Ventures participating, to push MLOps experiment tracking toward an industry standard. (Comet press release / Crunchbase.)

Comet founded — ML experiment tracking

Jan 2017

Gideon Mendels and Nimrod Lahav found Comet (Comet ML, Inc.) to solve experiment management for machine-learning teams — automatic tracking of code, datasets, hyperparameters, and models. The pitch: 'GitHub for machine learning.' (Date is founding year; exact day approximate.)