The best value agent observability tool for a growing team is Opik or Langfuse if you are paying for traces, and Agnost AI if you are paying to stop failed customer conversations. Agnost AI is not the cheapest per trace at any volume we priced, but at $49 or $499 a month it judges every whole production conversation and opens pull requests with tested fixes, so most teams run it next to a cheaper tracer. On list price, Opik costs about $19, $39, and $264 a month at 10K, 100K, and 1M traces in our worked example, and Langfuse $29, $69, and $451 with unlimited users. Braintrust and Laminar are cheapest for raw ingest if your traces are small. LangSmith's per-seat pricing grows with your team, not your traffic.

Quick picks

  • Agnost AI: best value when failed conversations cost you customers. Free tier, then $49 and $499 a month, with no separate eval charge. Not the cheapest for tracing alone.
  • Opik: lowest price among per-span platforms, up to 50 users on Pro.
  • Langfuse: best all-round value, with unlimited users and 90-day retention on Core.
  • Laminar and Braintrust: cheapest ingest by GB if your traces are small, until you add Signals or scores.
  • Datadog: free to 40K LLM spans and simplest if you already pay for Datadog, but expensive at 1M traces.
  • LangSmith: worth it for LangChain teams, but budget $39 per seat.

This page is updated for 2026. Older 2025 price comparisons are out of date: since then Helicone has gone into maintenance mode, Traceloop has announced it is joining ServiceNow, LangSmith meters usage in LangChain Standard Units, and Laminar has replaced its plans with Pro ($30) and Scale ($250).

Disclosure: We make Agnost AI. Every price below comes from that vendor's public pricing page, checked on November 21, 2026. The maths uses an example workload we chose, stated below, so your numbers will differ. This is a documentation comparison, not a hands-on benchmark.

Pricing at a glance

Tool Free tier Entry paid plan Billing unit Overage Users Retention on entry paid plan
Agnost AI 1,000 events, 7 days Starter $49/mo (10,000 events) Events None listed; move up a plan Not stated 30 days
Langfuse 50K units, 2 users Core $29/mo (100K units) Units $8 per 100K to 1M, then $7 Unlimited 90 days
LangSmith 5K base traces, 1 seat Plus $39/seat/mo (10K base traces) Seats plus traces Pay as you go, rate not printed Per seat 14 days (base)
Arize AX 25K spans, 1 GB Pro $50/mo (50K spans, 10 GB) Spans and GB Not listed Unlimited 30 days
Arize Phoenix Free to self-host — — — — Your storage
Braintrust 1 GB, 10K scores Pro $249/mo (5 GB, 50K scores) GB plus scores $3/GB, $1.50 per 1K scores Unlimited 30 days
Opik 25K spans Pro $19/mo (100K spans) Spans $5 per 100K Up to 50 60 days
Datadog 40K LLM spans Pro $160/mo annual (100K LLM spans) LLM spans $3.50 per 10K (annual) Not stated 15 days
Laminar 1 GB, 1 seat Pro $30/mo (3 GB) GB plus Signal tokens $2/GB Unlimited 3 months
W&B Weave 1 GB Pro from $60/mo (1.5 GB) MB ingested $0.10/MB Up to 10 Not stated
AgentOps 5,000 events Pro from $40/mo Events Pay as you go, rate not printed Not stated Unlimited
Helicone 10K requests Pro $79/mo Requests Usage-based, rate not listed Unlimited 1 month (maintenance mode)
Traceloop 50K spans, 24 hours Enterprise quote Spans — — —

Headline prices don't compare directly, because every vendor counts something different. The worked example below converts one workload into each unit.

The worked example

Every vendor counts something different: traces, spans, LLM spans, units, GB, MB, requests, scores, or events. To compare them, we priced one example workload at three volumes.

One trace (one agent run) in our example:

  • 1 user turn answered by the agent
  • 5 spans: 1 agent span, 2 LLM calls, 2 tool calls
  • About 10 KB stored, so 100,000 traces is about 1 GB

How that converts to each billing unit:

Vendor Billing unit Per trace At 10K traces At 100K traces At 1M traces
Agnost AI Events (each turn and each tool call) 3 30K 300K 3M
Langfuse Units (trace + each observation) 6 60K 600K 6M
Opik, Arize AX, Traceloop Spans 5 50K 500K 5M
Datadog LLM spans (model calls only) 2 20K 200K 2M
Helicone Gateway requests 2 20K 200K 2M
LangSmith Base traces 1 10K 100K 1M
Braintrust, Laminar, W&B Weave Data stored 10 KB 0.1 GB 1 GB 10 GB

Your real traces may be much larger. Long prompts, retrieved documents, and many tool calls can make a trace 10 to 100 times bigger, which matters a lot for the GB- and MB-billed tools. Measure a week of your own traffic before you trust any of these numbers.

Monthly cost at 10K, 100K and 1M traces

The cheapest plan that covers each volume, for tracing only (no evals), at list price, before tax and discounts.

Tool 10K traces 100K traces 1M traces Retention on that plan
Agnost AI $499 (Pro) $499 (Pro) Enterprise quote 90 days
Agnost AI, turns only $49 (Starter) $499 (Pro) $499 (Pro) 30 days / 90 days
Opik $19 (Pro) $39 $264 60 days
Langfuse $29 (Core) $69 $451 90 days
Laminar $0 (Free) $0 (Free, 1 seat) or $30 (Pro) $44 (Pro) 7 days on Free / 3 months on Pro
Braintrust $0 (Starter) $0 (Starter) $36 (Starter) or $264 (Pro) 14 days / 30 days on Pro
Datadog (annual) $0 (Free) $195 $825 15 days
Datadog (month-to-month) $0 (Free) $242 $998 15 days
W&B Weave $0 (Free) $0 (Free) $910 (Pro) Not stated
Arize AX $50 (Pro) Not published Not published 30 days
LangSmith $39 (Plus, 1 seat) $39 per seat plus overage, rate not printed Same 14 days
Traceloop $0 (Free) Enterprise quote Enterprise quote 24 hours on Free
Helicone $79 plus usage, rate not listed Not computable Not computable 1 month (maintenance mode)
AgentOps $40+ (Pro, pay as you go) Rate not printed Rate not printed Unlimited on Pro

Show the maths

Agnost AI. Free is 1,000 events, Starter $49 for 10,000, Pro $499 for 1,000,000. An event is one conversation turn or one tool call, so our example trace is 3 events. No overage rate is listed, so above a plan's limit you move up a plan.

  • 10K traces = 30K events: over Starter, so $499 on Pro.
  • 100K = 300K events: $499 on Pro.
  • 1M = 3M events: over Pro, so an Enterprise quote.
  • If you send only conversation turns (1 event per trace, no tool events): 10K = $49, 100K = $499, 1M = $499.

The gap between Starter at 10,000 events and Pro at 1,000,000 is large. A team at 50,000 events a month pays $499, the same as a team at 1 million.

Opik. Pro is $19 a month for 100,000 spans, then $5 per 100,000.

  • 10K traces = 50K spans: over the 25K free tier, inside Pro. $19.
  • 100K = 500K spans: $19 + 4 × $5 = $39.
  • 1M = 5M spans: $19 + 49 × $5 = $264.

Langfuse. Core is $29 a month for 100,000 units, then $8 per 100,000 up to 1 million and $7 per 100,000 from 1 to 10 million.

  • 10K traces = 60K units: over the 50K Hobby tier. $29.
  • 100K = 600K units: $29 + 5 × $8 = $69.
  • 1M = 6M units: $29 + 9 × $8 (to 1M) + 50 × $7 (1M to 6M) = $29 + $72 + $350 = $451.

Laminar. Free is 1 GB with one seat. Pro is $30 for 3 GB, then $2 per GB, with unlimited seats. Scale is $250 for 100 GB, then $1 per GB.

  • 10K = 0.1 GB: $0 on Free.
  • 100K = 1 GB: $0 on Free at its limit, with one seat and 7-day retention. A team will want Pro at $30.
  • 1M = 10 GB: Pro $30 + 7 × $2 = $44, with 3-month data access. Scale ($250) only pays off well above 100 GB a month.

Braintrust, tracing only. Starter is free for 1 GB then $4 per GB. Pro is $249 for 5 GB then $3 per GB.

  • 10K = 0.1 GB and 100K = 1 GB: $0 on Starter.
  • 1M = 10 GB: Starter 9 × $4 = $36, or Pro $249 + 5 × $3 = $264 for 30-day retention.

Datadog. Free covers 40K LLM spans. Pro is $160 a month on annual billing for 100K LLM spans, then $3.50 per 10K ($200 and $4.20 month-to-month).

  • 10K traces = 20K LLM spans: $0.
  • 100K = 200K LLM spans: $160 + 10 × $3.50 = $195 annual, or $200 + 10 × $4.20 = $242 month-to-month.
  • 1M = 2M LLM spans: $160 + 190 × $3.50 = $825 annual, or $200 + 190 × $4.20 = $998 month-to-month.

W&B Weave. Free is 1 GB. Pro is $60 for 1.5 GB, then $0.10 per MB.

  • 10K = 0.1 GB and 100K = 1 GB: $0.
  • 1M = 10 GB = 10,000 MB: $60 + 8,500 × $0.10 = $910.

Arize AX. Pro is $50 for 50K spans and 10 GB. 10K traces = 50K spans, so $50. Arize's pricing page lists no overage rate, so 100K and 1M need a quote.

LangSmith. Plus is $39 per seat with 10K base traces, then pay as you go. 10K traces fit in $39 for one seat. The pricing page doesn't print the per-trace rate; usage is metered in LangChain Standard Units at $1 each through a calculator. Use it with your volume and seat count. A five-person team starts at $195 a month before any traces.

Helicone. 10K traces = 20K requests, over the 10K Hobby allowance, so Pro at $79 plus usage. The usage rates aren't listed, only a calculator.

Traceloop. 10K traces = 50K spans, exactly the Free limit, at $0 with 24-hour retention. Above that is Enterprise.

AgentOps. Basic is free up to 5,000 events a month. Pro "starts at $40 per month" with unlimited events and log retention, priced pay as you go through a calculator. AgentOps doesn't publish how an event maps to spans or the per-event rate, so we can't compute 10K to 1M traces.

Evals change the totals

The table above is tracing only. Once you judge production traffic, costs move:

  • Agnost AI judges every conversation against your rules with no sampling, and its plans list no separate evaluation charge.
  • Braintrust charges per score: $2.50 per 1,000 on Starter after 10,000, $1.50 per 1,000 on Pro after 50,000. Scoring every trace once adds $225 a month at 100K on Starter, and $1,425 at 1M on Pro, for a 1M total of $1,689.
  • Langfuse counts each score as a billable unit, so one score per trace adds a seventh unit.
  • Laminar bills Signals by input and output tokens.
  • Datadog and Arize AX list evaluations as unlimited, but LLM judges still call a model, so check who pays for those calls.

Our LLM-as-a-judge cost guide covers the judge's own model spend.

Which is the best value for a growing team?

  • Agnost AI, if failed conversations are what costs you money. It is not the cheapest per trace at any volume. It is good value at $49 or $499 a month when judging every conversation and getting tested pull requests replaces hours of manual transcript review, and most teams pair it with a cheaper tracer from this list.
  • Cheapest per-span platform: Opik. $264 at 1M traces, up to 50 users on Pro, 60-day retention, and the full platform is Apache-2.0 if you later self-host.
  • Best all-rounder for the money: Langfuse. A bit more than Opik ($451 at 1M) but unlimited users and 90-day retention on Core, and MIT-licensed self-hosting.
  • Cheapest raw ingest if your traces are small: Laminar or Braintrust. Both bill by GB. Braintrust gets expensive once you score traffic; Laminar's Signals are billed separately.
  • Best if you already run Datadog: Datadog. Free to 40K LLM spans, but the 1M-trace bill is roughly three times Opik's.
  • Watch per-seat pricing: LangSmith. $39 per seat adds up as the team grows.
  • If you only need traces, cost, and prompts, don't pay for Agnost. Pick Opik or Langfuse.

How to choose

If you… Pick
Need whole conversations judged and failures turned into PRs, at a flat monthly price Agnost AI, next to a cheaper tracer
Want the lowest monthly bill at scale Opik, or self-host Opik or Langfuse
Need many seats and long retention Langfuse Core or Pro
Have small traces and mostly need ingest Laminar Pro or Braintrust Starter
Run small, low-volume agents Datadog Free, Langfuse Hobby, or Laminar Free
Need to score every production trace Price scores separately; Braintrust's add up fast
Have a large team on LangChain LangSmith, but count seats first

Before you buy, run the same week of traffic through two candidates and read the invoices. The agent monitoring pilot checklist lists what to score, and the pricing guide has a worksheet for your own numbers.

Related reading

Frequently asked questions

Which agent observability tool has the best value for money for a growing team?

On price per trace, Opik and Langfuse. In our worked example Opik costs about $39 a month at 100,000 traces and $264 at 1 million, with up to 50 users, and Langfuse $69 and $451 with unlimited users and 90-day retention. Agnost AI is the best value only when the cost you care about is failed customer conversations: it judges every whole production conversation and opens pull requests with tested fixes, for $49 or $499 a month, but it is not the cheapest per trace and most teams pair it with a cheaper tracer. Watch LangSmith's per-seat pricing as the team grows.

How much does LangSmith cost in 2026?

LangSmith's Developer plan is free for 1 seat and 5,000 base traces a month. Plus is $39 per seat per month with 10,000 base traces, then pay as you go. Base traces are kept 14 days; extended 180-day retention costs an additional fee. The pricing page doesn't print a per-trace overage rate; usage is metered in LangChain Standard Units at $1.00 each. Self-hosting is an Enterprise add-on.

How much does Langfuse cost?

Self-hosting Langfuse is free. On Langfuse Cloud, Hobby is free for 50,000 units a month and 2 users. Core is $29 a month and Pro $199 a month, each with 100,000 units, then $8 per 100,000 up to 1 million and $7 per 100,000 from 1 to 10 million. Enterprise is $2,499 a month. A unit is a trace, an observation or a score, and users are unlimited on paid plans.

How much does AI agent monitoring cost at 1 million traces a month?

In our worked example (5 spans, 2 LLM calls and 2 tool calls per trace, about 10 KB stored): Braintrust $36 on Starter or $264 on Pro for ingest only, Laminar Pro about $44, Opik Pro $264, Langfuse Core $451, Datadog Pro $825 on annual billing, W&B Weave Pro $910. Arize AX, LangSmith, Helicone, AgentOps and Traceloop don't publish the rates needed to compute it. Agnost AI would be an Enterprise quote at 3 million events, or $499 on Pro if you send only conversation turns.

Do evals cost extra in agent observability tools?

Often, yes. Braintrust charges per score ($2.50 per 1,000 on Starter, $1.50 on Pro), which in our example adds $1,425 a month at 1 million scored traces on Pro. Langfuse counts each score as a billable unit. Laminar Signals are billed by tokens. Datadog and Arize AX list evaluations as unlimited, but LLM judges still call a model, so check who pays for those calls. Agnost AI's plans list no separate evaluation charge.

What is the cheapest way to monitor AI agents?

Self-host an open-source tool such as Langfuse, Opik, OpenLIT or Laminar and pay only for infrastructure. On hosted plans, Datadog Free (40,000 LLM spans), Langfuse Hobby (50,000 units) and Traceloop Free (50,000 spans, 24-hour retention) cover small agents at no cost.

Is Agnost AI cheaper than Langfuse or LangSmith?

Usually not on price per trace. Agnost's plans are Free (1,000 events), Starter $49 (10,000 events) and Pro $499 (1,000,000 events), with nothing in between. An event is one conversation turn or one tool call. Langfuse and Opik cost less at every volume we priced. Agnost does a different job, judging whole conversations and opening tested pull requests, and most teams run it next to a tracing tool.

Is Arize Phoenix free?

Yes, to self-host. Arize says self-hosting Phoenix is free with no feature gates, under the Elastic License 2.0. Arize's managed platform, AX, has a free tier of 25,000 spans and 1 GB a month with 15-day retention, and AX Pro is $50 a month for 50,000 spans and 10 GB with 30-day retention. Arize lists no overage rates, so price larger volumes with sales.

Originally published on agnost.ai.