Logging 1 million agent traces a month (4 spans and about 10 KB each) costs $36 on Braintrust Starter, $44 on Laminar Starter, $214 on Opik Pro, $311 on Langfuse Core and $5,067 on LangSmith Plus for 3 engineers, as of October 30, 2026 (our arithmetic on list rates). Billing units drive the gap: per GB is cheapest for small traces, per trace is dearest. Self-hosted Langfuse or Phoenix costs nothing in licences.
Plan fees, overage rates and retention from nine tracing and eval platforms, priced on one agent workload at 1 million and 10 million traces a month.
LLM observability bills look similar on the plan cards, all within a few hundred dollars a month, and then diverge by two orders of magnitude once traffic arrives. The deciding factor is the unit each vendor meters: gigabytes of logged data, spans, Langfuse's units, or whole traces, and how many of them one agent run produces.
Toolradar data: our AI observability category lists 54 tools, and 61% of them offer a free or freemium plan, often a capped cloud tier next to an open-source build you can host yourself. The AI observability statistics page covers the wider category.
Methodology: we read the published plan and overage rates of nine platforms on each vendor's own pricing page and docs as of October 30, 2026, in US dollars. Every monthly total is our arithmetic on one workload: 1 million agent traces a month, each with a root span plus 3 LLM-call spans (4 spans, or 4 Langfuse units), about 10 KB of logged prompts and outputs per trace (10 GB a month), and 3 engineers with seats. Totals exclude evals, scores, model credits, extra retention and tax unless stated. Ranking is by cost fit for that team; no vendor paid for a position.
Top Picks
Selected by pricing one agent workload (1 million traces a month, 4 spans and about 10 KB each, 3 engineers) on every LLM observability platform's published plan and overage rates, then comparing retention and the self-hosting option. How we rate
Open-source LLM engineering platform for tracing, evals, prompt management and metrics, cloud or self-hosted.
Starting price
Core $29/mo, unlimited users
Best for
Most teams shipping an agent that want unlimited users and a self-host exit.
Evals and observability platform billed on processed data and scores, with unlimited users.
Starting price
Starter $0 + $4/GB
Best for
Eval-heavy teams with compact traces that want no platform fee to start.
Open-source observability for AI agents, billed on GB of trace data with unlimited seats on paid plans.
Starting price
Starter $30/mo, 3 GB included
Best for
Small agent teams that want a low flat fee and 30 days or more of history.
| Tool | Starting price | Rating | Best for |
|---|---|---|---|
| Langfuse Visit Langfuse website | Core $29/mo, unlimited users | n/a | Best for: Most teams shipping an agent that want unlimited users and a self-host exit. |
| Braintrust Visit Braintrust website | Starter $0 + $4/GB | n/a | Best for: Eval-heavy teams with compact traces that want no platform fee to start. |
| Laminar Visit Laminar website | Starter $30/mo, 3 GB included | n/a | Best for: Small agent teams that want a low flat fee and 30 days or more of history. |
| Opik by Comet Visit Opik by Comet website | Pro $19/mo, 100k spans | n/a | Best for: Teams that want the cheapest span-based cloud plan and an open-source fallback. |
| Arize Phoenix Visit Arize Phoenix website | Free, self-hosted | n/a | Best for: Teams that want zero licence cost and can run their own tracing backend. |
| Portkey Visit Portkey website | Production $49/mo, 100k logs | 4.6 18 reviews | Best for: Teams that want routing, fallbacks and caching on the same bill as observability. |
| Helicone Visit Helicone website | Pro $79/mo + usage | n/a | Best for: Teams that want proxy-based logging with a one-line integration and unlimited seats. |
| LangSmith Visit LangSmith website | Plus $39/seat/mo + traces | n/a | Best for: LangChain and LangGraph teams that also deploy agents through LangSmith. |
| Arize AX Visit Arize AX website | Pro $50/mo, 50k spans | 4.3 48 reviews | Best for: Enterprises that want a managed platform with custom volume and self-hosted options. |
Langfuse
Open-source LLM engineering platform for tracing, evals, prompt management and metrics, cloud or self-hosted.
- Starting price
- Core $29/mo, unlimited users
- Free plan
- Yes
Best for: Most teams shipping an agent that want unlimited users and a self-host exit.
- Pro: Overage falls from $8 to $6 per 100k units as volume grows.
- Pro: Pro at $199 a month keeps data 3 years.
- Pro: Self-hosted version is free under the MIT license.
- Con: Every observation and score counts as a unit, so long agents add up.
- Con: SSO and fine-grained RBAC need the $300 a month Teams add-on.
Langfuse's pricing is quite generous, with a free Hobby tier offering 50k units per month, 30 days of data access and 2 users.
Watch out
Usage beyond included units at $8/100k units (graduated, lower with volume)
Braintrust
Evals and observability platform billed on processed data and scores, with unlimited users.
- Starting price
- Starter $0 + $4/GB
- Free plan
- Yes
Best for: Eval-heavy teams with compact traces that want no platform fee to start.
- Pro: Starter has no platform fee and includes 1 GB and 10k scores.
- Pro: Unlimited users, projects and experiments on every plan.
- Pro: 6 to 12 months of Pro free for qualifying startups.
- Con: Starter keeps data only 14 days.
- Con: Pro carries a large platform fee before any usage.
Braintrust's free tier is generous for small projects, but the Pro plan's overage charges on processed data and scores can escalate quickly for high-volume teams.
Watch out
1 GB processed data, then $4 per GB (Starter)
Laminar
Open-source observability for AI agents, billed on GB of trace data with unlimited seats on paid plans.
- Starting price
- Starter $30/mo, 3 GB included
- Free plan
- Yes
Best for: Small agent teams that want a low flat fee and 30 days or more of history.
- Pro: Overage of $2 per GB on Starter and $1.50 on Pro.
- Pro: Pro keeps 6 months of data with 10 GB included.
- Pro: Unlimited projects and seats from Starter.
- Con: Free plan keeps data 7 days and allows 1 seat.
- Con: Signals analysis is billed on top of the plan.
- Starting price
- Pro $19/mo, 100k spans
- Free plan
- Free to use
Best for: Teams that want the cheapest span-based cloud plan and an open-source fallback.
- Pro: Extra spans cost $5 per 100k.
- Pro: Up to 50 team members on Pro.
- Pro: Open-source build shares the hosted codebase.
- Con: Keeping spans 400 instead of 60 days costs $29 per 100k spans.
- Con: Span billing grows with agent depth, not just traffic.
Opik by Comet's pricing is extremely generous, offering a fully-featured Free tier that includes advanced capabilities like automated agent optimization and built-in guardrails.
Watch out
Potential overage fees for high usage
Arize Phoenix
Open-source, local-first tracing and evaluation tool from Arize, built on OpenTelemetry.
- Starting price
- Free, self-hosted
- Free plan
- Yes
Best for: Teams that want zero licence cost and can run their own tracing backend.
- Pro: Runs on a laptop, in a container or in your cloud.
- Pro: Same OpenTelemetry instrumentation as Arize AX, so moving to managed later is direct.
- Pro: Tracing, evals, experiments and prompt iteration in one tool.
- Con: You run the storage, upgrades and backups yourself.
- Con: Managed hosting and Arize AX features need a paid AX plan.
Phoenix offers a very generous free tier with its open-source solution and a substantial free tier for AX, making it highly accessible.
Watch out
Overage fees for exceeding trace spans/ingestion
- Starting price
- Production $49/mo, 100k logs
- Free plan
- Yes
- Rating
- 4.6 (18 reviews)
Best for: Teams that want routing, fallbacks and caching on the same bill as observability.
- Pro: Extra logs cost $9 per 100k requests.
- Pro: Fallbacks, load balancing and retries included.
- Pro: On the free plan, requests beyond the log limit still go through; only logging stops.
- Con: Logs are kept 30 days on Production.
- Con: Free plan keeps logs only 3 days.
Portkey's pricing is fair, offering a generous free Developer tier with 10k requests/month.
- Starting price
- Pro $79/mo + usage
- Free plan
- Yes
Best for: Teams that want proxy-based logging with a one-line integration and unlimited seats.
- Pro: Unlimited seats on Pro.
- Pro: Free Hobby plan with 10,000 requests a month.
- Pro: 50% off the first year for startups under 2 years old.
- Con: Usage rates appear only in the pricing calculator.
- Con: Team plan at $799 a month is the step for SOC 2 and HIPAA.
It's best for developers and small to medium-sized businesses building and scaling AI applications.
Watch out
Potential overage fees if exceeding Pro tier limits
LangSmith
LangChain's platform for tracing, evals, deployment and agent tooling, priced per seat plus usage.
- Starting price
- Plus $39/seat/mo + traces
- Free plan
- Yes
Best for: LangChain and LangGraph teams that also deploy agents through LangSmith.
- Pro: Plus includes one free small serverless deployment.
- Pro: Extended traces keep data 180 days.
- Pro: Up to $10,000 in credits for VC-backed startups.
- Con: Per-seat fee plus per-trace usage make it the costliest option at volume.
- Con: Base traces are kept only 14 days.
Plus is $39/seat/month with 10,000 traces and one free small serverless deploy.
Watch out
Traces over the included 5k/10k are pay-as-you-go. Seats are only the floor.
- Starting price
- Pro $50/mo, 50k spans
- Free plan
- Yes
- Rating
- 4.3 (48 reviews)
Best for: Enterprises that want a managed platform with custom volume and self-hosted options.
- Pro: Unlimited users and evals on every plan.
- Pro: Pro includes 10 GB of storage.
- Pro: Enterprise can be SaaS or self-hosted.
- Con: No published overage: production volume means an Enterprise quote.
- Con: Pro keeps data 30 days.
Arize AI offers a generous free tier (AX Free) and an open-source option (Phoenix), making it highly accessible for individual developers and small teams.
Watch out
Self-hosting is an Enterprise add-on
Other AI Observability tools worth considering
More published tools from our AI Observability category, ordered by our category ranking. They are not part of the editorial picks above.
How LLM observability pricing works
LLM observability tools record each request your application makes to a model, group them into traces (one agent run, one chat turn), and add evals, prompt management and dashboards on top. Vendors meter that logging in five different units:
| Billing unit | What counts | Vendors |
|---|---|---|
| Data volume (GB) | Bytes of prompts, outputs and metadata ingested | Braintrust, Laminar, W&B Weave |
| Spans | Every step inside a trace | Opik, Arize AX |
| Units | Traces plus observations plus scores | Langfuse |
| Traces, plus seats | One per application run, however many steps | LangSmith |
| Requests or logs | Each LLM call through a gateway or proxy | Portkey, Helicone |
Why the unit decides the bill. An agent that makes 3 model calls per run produces 1 trace, 4 spans and 4 Langfuse units. A GB-metered tool charges the same whether that data arrives as 1 trace or 40 spans, so short spans are cheap there; a per-trace tool charges the same for a 1-step and a 40-step agent, so long agents are cheap there.
Retention is the hidden second price. The default window ranges from 7 days on Laminar's free plan to 3 years on Langfuse Pro. Keeping data longer costs extra on several plans, so check retention before you compare monthly fees.
How much does logging 1 million traces a month cost?
One workload, priced on every platform: 1 million traces a month, 4 spans and about 10 KB each (10 GB), 3 engineers. Rates as of October 30, 2026; totals are our arithmetic.
| Platform and plan | Arithmetic | Monthly total | Default retention |
|---|---|---|---|
| Braintrust Starter | 9 GB over the free 1 GB at $4 | $36 | 14 days |
| Laminar Starter | $30 + 7 GB over 3 GB at $2 | $44 | 30 days |
| Laminar Pro | 10 GB included | $150 | 6 months |
| Opik Pro | $19 + 39 x $5 per extra 100k spans | $214 | 60 days |
| Braintrust Pro | $249 + 5 GB over 5 GB at $3 | $264 | 30 days |
| Portkey Production (3M LLM calls) | $49 + 29 x $9 per extra 100k | $310 | 30 days for logs |
| Langfuse Core (4M units) | $29 + $72 (100k to 1M) + $210 (1M to 4M) | $311 | 90 days |
| Langfuse Pro (4M units) | $199 + the same $282 of usage | $481 | 3 years |
| W&B Weave Pro | $60 + 8,500 MB over 1.5 GB at $0.10 | $910 | Not stated on the plan card |
| LangSmith Plus, 3 seats | $117 + 990,000 traces at 0.005 LSU ($1 each) | $5,067 | 14 days on base traces |
| Arize AX Pro | 4M spans exceeds the 50k included | Enterprise quote | 30 days |
| Helicone Pro | $79 + usage on 3M requests | Check Helicone's calculator | 1 month |
At 10 million traces the order holds but the gaps widen. Our arithmetic gives $396 on Braintrust Starter, $224 on Laminar Starter, $285 on Laminar Pro, $2,014 on Opik Pro, $2,681 on Langfuse Core and $50,067 on LangSmith Plus. Langfuse's graduated rate falls to $6.50 per 100k units between 10M and 50M, which softens but does not close the gap with GB-metered tools.
What the table hides. GB-metered bills move with trace size: agents that log full documents or images can triple the 10 KB assumption. Evals cost extra on several platforms (Braintrust scores beyond the included 10k are $2.50 per 1,000 on Starter), and retention beyond the default is a separate line.
Key Features to Look For
Billing unit that fits your traces (Essential)
Short traces with many small spans are cheapest on GB-metered plans (Braintrust, Laminar). Long agent runs logged as one trace suit per-trace billing, but LangSmith's per-trace rate makes high volume expensive.
Retention window (Essential)
Defaults run from 14 days (Braintrust Starter, LangSmith base traces) to 3 years (Langfuse Pro). Braintrust Pro adds $0.50 per GB a month for longer retention and Opik charges extra per 100k spans to keep them 400 days.
Seat pricing (Essential)
LangSmith Plus charges per seat on top of usage. Langfuse Core, Braintrust, Laminar Starter and Helicone Pro include unlimited users, which matters more than the base fee once a whole team needs access.
Self-hosting (Important)
Langfuse is free to self-host under the MIT license, Arize Phoenix is open source and self-hosted, and Opik's open-source build is the same codebase as its cloud. You pay only for the servers.
Graduated volume pricing (Important)
Langfuse's overage drops from $8 to $7, $6.50 and $6 per 100k units as volume grows. Laminar's overage falls from $2 per GB on Starter to $1.50 on Pro.
Gateway features (Nice to have)
Portkey and Helicone sit in the request path, so the same plan adds fallbacks, retries, load balancing and caching. Logging-only tools need a separate gateway for that.
What to measure before you pick an LLM observability tool
-
Count spans per trace on a real agent run: a 3-call agent is 4 spans, a tool-heavy agent can be 40, and span or unit pricing scales with that number.
-
Measure logged bytes per trace for a day. GB-metered quotes are only as good as that average.
-
Count who needs access. Per-seat pricing on a 10-person AI team adds $390 a month on LangSmith Plus before any trace.
-
Decide how long you need traces: 14 days covers debugging, but evals against last quarter's production data need months.
-
Check whether evals and LLM-as-judge scores are billed separately from tracing.
-
If data residency or privacy rules apply, price self-hosting Langfuse, Phoenix or Opik before any cloud plan.
Evaluation Checklist
Instrument one service with OpenTelemetry and send the same traces to two vendors for a week.
Read each vendor's usage page after a day and multiply by 30, rather than trusting the calculator.
Open a trace older than the plan's retention window to confirm what happens to it.
Run one LLM-as-judge eval and check how it appears on the invoice preview.
Set a usage limit or alert so a logging loop cannot run up overage.
Export a dataset of traces to check you can leave with your data.
Pricing Overview
Free and self-hosted
$0
Low-fee usage plans
$0 to $49 a month base
Team plans
Longer retention, RBAC and support: Arize AX Pro, Helicone Pro, Laminar Pro, Langfuse Pro, Braintrust Pro.
$50 to $249 a month base
Seat-based
LangChain and LangGraph teams that want LangSmith's deployment and agent tools alongside tracing.
$39 per seat a month plus traces
Enterprise
SSO, audit logs, SLAs and compliance: Helicone Team, Langfuse Enterprise, Arize AX Enterprise, Braintrust Enterprise.
$799 to $2,499 a month, or custom
Pricing Comparison
| Platform | Free tier | Entry paid plan | Overage | Retention on entry plan | 1M traces a month (our workload) |
|---|---|---|---|---|---|
Langfuse | Hobby: 50k units, 2 users | Core $29/month, unlimited users | $8 per 100k units, falling to $6 with volume | 90 days | $311 |
Developer: 1 seat, 5k base traces | Plus $39 per seat/month, 10k traces | 0.005 LSU per extra trace; 1 LSU = $1 | 14 days base, 180 days extended | $5,067 with 3 seats | |
Braintrust | Starter $0: 1 GB, 10k scores | Pro $249/month: 5 GB, 50k scores | $4/GB on Starter, $3/GB on Pro | 14 days Starter, 30 days Pro | $36 on Starter |
Laminar | Free: 1 GB, 7 days, 1 seat | Starter $30/month: 3 GB, unlimited seats | $2/GB on Starter, $1.50/GB on Pro | 30 days Starter, 6 months Pro | $44 on Starter |
Opik (Comet) | Free Cloud: 25k spans, 10 members | Pro $19/month: 100k spans, 50 members | $5 per 100k spans | 60 days | $214 |
Portkey | Free: 10k logs, 3-day retention | Production $49/month: 100k logs | $9 per 100k requests | 30 days for logs | $310 for 3M LLM calls |
Helicone | Hobby: 10k requests, 1 GB, 1 seat | Pro $79/month, unlimited seats | Usage-based on requests and storage | 1 month | Check the current rates on Helicone's site |
Free: 25k spans, 1 GB | Pro $50/month: 50k spans, 10 GB | No published overage; Enterprise above | 30 days | Enterprise quote | |
W&B Weave | Free: 1 GB ingestion | Pro from $60/month: 1.5 GB ingestion | $0.10 per MB | Not stated on the plan card | $910 |
Prices as of October 30, 2026 from Langfuse pricing and Langfuse billable units docs, LangSmith pricing, Braintrust pricing, Laminar pricing, Opik and Comet pricing, Portkey pricing, Helicone pricing, Arize pricing and Weights & Biases pricing. The last column is our arithmetic for 1 million traces a month with 4 spans and about 10 KB each (4M Langfuse units, 4M spans, 10 GB, 3M LLM calls) and 3 engineers, before evals, model credits and extra retention. LangSmith's per-trace rate is the one its usage calculator applies to traces beyond the included 10k.
Mistakes to Avoid
- ×
Comparing plan fees instead of overage. A per-seat plan looks cheaper than a large platform fee until 1 million traces arrive and per-trace billing adds thousands.
- ×
Counting traces when the vendor bills spans or units. A 3-call agent run is 1 trace but 4 Langfuse units and 4 Opik spans.
- ×
Logging full retrieved documents into every span on a GB-metered plan, which multiplies the data line.
- ×
Setting LangSmith's default retention to extended traces for everything: extended traces cost more, so the same budget covers fewer of them.
- ×
Picking Arize AX Pro for production volume: its 50k included spans cover a prototype, not 1 million runs.
Expert Tips
- →
Sample before you pay: logging 10% of production traces plus every error trace cuts logged volume by about 90% while keeping the failures you debug.
- →
Run evals offline on a dataset rather than online on every trace if your vendor bills scores separately.
- →
Start on self-hosted Langfuse or Phoenix while volume is unknown, then move to the cloud plan once you can measure units or GB per day.
- →
Compare the leaders head to head: Langfuse vs LangSmith, Braintrust vs LangSmith, LangSmith alternatives and Langfuse alternatives.
- →
For a ranking by features rather than cost, see the best LLM observability tools and the best LLM monitoring tools guides.
Red Flags to Watch For
-
A pricing page that shows only a calculator for usage, with no per-unit rate printed.
-
Per-trace pricing with no cap on spans or bytes per trace stated next to it.
-
A low entry plan with no published overage rate, so growth means a sales call.
-
Retention shorter than the eval cycle you plan to run on production data.
The Bottom Line
For most teams shipping an agent, Langfuse is the balanced pick: a low Core fee with unlimited users, 90-day retention, published volume rates and a free MIT-licensed self-hosted version if the cloud bill grows. Braintrust and Laminar produce the smallest bills for compact traces because they meter gigabytes, and Opik is the cheapest span-metered option.
LangSmith is the natural fit for LangChain and LangGraph teams that also use its deployment tools, but per-seat fees plus per-trace usage make it the most expensive way to log high volume. Gateway-first teams should price Portkey or Helicone, which add routing and caching to the same plan. Browse every option in the AI observability category, or the AI agent observability guide for agent-specific tracing.
Frequently Asked Questions
How much does LLM observability cost per month?
For 1 million agent traces a month (4 spans and about 10 KB each) and 3 engineers, list rates as of October 30, 2026 give about $36 on Braintrust Starter, $44 on Laminar Starter, $214 on Opik Pro, $311 on Langfuse Core and $5,067 on LangSmith Plus (our arithmetic). Small apps stay inside free tiers or the base fee.
Is Langfuse cheaper than LangSmith?
At volume, yes. For our 1-million-trace workload Langfuse Core costs about a sixteenth of LangSmith Plus with 3 seats (our arithmetic), because LangSmith adds a seat fee and 0.005 LSU ($0.005) per trace beyond 10k. Langfuse is also free to self-host under the MIT license.
Is there a free LLM observability tool?
Yes. Langfuse Hobby includes 50k units a month, Opik Free Cloud 25k spans, Arize AX Free 25k spans and Braintrust Starter 1 GB with no platform fee. For unlimited volume at no licence cost, self-host Langfuse, Arize Phoenix or Opik's open-source build.
How much does Helicone cost?
Helicone's Hobby plan is free for 10,000 requests, 1 GB of storage and 1 seat. Pro costs $79 a month with unlimited seats and Team $799 a month with 5 organizations and SOC 2 and HIPAA compliance; both add usage-based charges on requests and storage beyond the free allowance.
What does 10 million traces a month cost?
With 4 spans and about 10 KB per trace, our arithmetic on October 2026 list rates gives about $224 on Laminar Starter, $396 on Braintrust Starter, $2,014 on Opik Pro, $2,681 on Langfuse Core and $50,067 on LangSmith Plus with 3 seats. At that volume, ask for annual commitment pricing.
How long do LLM observability tools keep traces?
It depends on the plan: 14 days on Braintrust Starter and LangSmith base traces, 30 days on Laminar Starter and Arize AX Pro, 60 days on Opik, 90 days on Langfuse Core and 3 years on Langfuse Pro. Longer retention costs extra on Braintrust Pro, Opik and LangSmith.
Is self-hosting LLM observability worth it?
For high volume or strict data rules, often. Self-hosted Langfuse has no usage billing, and Phoenix and Opik run on your own infrastructure too. You trade the subscription for servers, storage and upkeep, which usually pays off once a cloud bill reaches several hundred dollars a month.
Cite this page: Toolradar, "LLM Observability Pricing Compared (2026)", updated November 2026, https://toolradar.com/guides/llm-observability-pricing-compared
Spotted an outdated price or a wrong fact? Suggest a correction: we check it against the vendor's own page. Work for one of these vendors? Claim your listing to update it yourself.
Related Guides
Ready to Choose?
Compare features, read reviews, and find the right tool.
Some offers on this page may be paid placements or contain affiliate links.