LangSmith vs Langfuse vs Braintrust Pricing: 2026 Cost Breakdown

Last updated October 13, 2026. Every price below was checked on the vendor’s own pricing page and billing docs on that date.

LangSmith pricing looks cheap at $39 per seat until trace volume arrives, Langfuse looks cheap until you count every span as a unit, and Braintrust looks expensive at $249 until you notice it barely moves with volume. The three vendors meter different things, so their list prices cannot be compared side by side without a workload model.

This breakdown builds that model. We define one agent run, price it on each vendor’s October 2026 list rates, and show the monthly bill at pilot, growth, and scale volumes. Placement is unpaid, per our editorial standards and disclosure policy.

If you are still choosing on features, read the Langfuse vs LangSmith vs Braintrust comparison first. This page answers the narrower question finance asks next: what will it cost us each month?

LangSmith vs Langfuse vs Braintrust Pricing at a Glance

The table lists each plan’s published price and what it includes. Read the meter column before the price column, because it decides how the bill grows.

Vendor and plan	Monthly price	Included usage	Overage	Users	Data retention
Langfuse Hobby	$0	50k units	None (hard cap)	2	30 days
Langfuse Core	$29	100k units	$8 per 100k units, falling to $6	Unlimited	90 days
Langfuse Pro	$199	100k units	Same graduated rates	Unlimited	3 years
Langfuse Enterprise	$2,499	100k units	Same, or custom yearly	Unlimited	3 years
LangSmith Developer	$0	5k base traces	$5 per 1,000 base traces	1 seat	14 days base
LangSmith Plus	$39 per seat	10k base traces	$5 per 1,000 base traces	Unlimited, paid	14 days base, 180 days extended
LangSmith Enterprise	Custom	Custom	Custom	Custom	Custom
Braintrust Starter	$0	1 GB, 10k scores	$4 per GB, $2.50 per 1,000 scores	Unlimited	14 days
Braintrust Pro	$249	5 GB, 50k scores, $100 model credits	$3 per GB, $1.50 per 1,000 scores	Unlimited	30 days, up to 180 at $0.50 per GB
Braintrust Enterprise	Custom	Custom	Custom	Unlimited	Custom

What Each Vendor Actually Meters

The single biggest pricing mistake is treating a Langfuse unit, a LangSmith trace, and a Braintrust gigabyte as the same thing. They measure different layers of the same agent run.

• Langfuse bills units. A unit is any tracing data point: the trace itself, each observation (span, event, or generation), and each score, per the billable units docs. An agent run with eight steps and no score is nine units.
• LangSmith bills traces and seats. One trace covers a full run no matter how many steps it holds. Beyond the included allowance each base trace costs 0.005 LangChain Standard Units (LSU), and 1 LSU equals $1, per the LangSmith pricing page.
• Braintrust bills processed data and scores. Payload size in gigabytes drives the bill, along with the count of LLM-as-judge, autoeval, or code scores, per Braintrust pricing.

A deep agent with 30 tool calls per run therefore costs about 3x more on Langfuse than a shallow one, costs the same on LangSmith, and costs more on Braintrust only if those calls carry large payloads.

LangSmith Pricing in 2026

LangSmith has three plans. Developer is free for one seat with 5,000 base traces per month. Plus costs $39 per seat per month with 10,000 base traces shared across the organization, and Enterprise is custom with self-hosted and hybrid deployment, per the pricing page.

The per-trace rate sits in a tooltip rather than the plan cards: 0.005 LSU per base trace, which is $5 per 1,000 traces, plus 0.0025 LSU to upgrade a trace to extended retention. An extended trace therefore costs $7.50 per 1,000 in total.

Base vs extended traces

Base traces keep 14 days. Extended traces keep 180 days for new SaaS traces from September 14, 2026, per the usage and billing docs. Some features, such as adding feedback or annotation, upgrade a trace to extended automatically, so a team that labels heavily pays the higher rate on more traces than it planned.

Other LangSmith line items

• Deployment: Plus includes one free small serverless deployment; extra runtime is metered in LSU per vCPU-hour and GiB-hour.
• Engine: the automated failure-analysis service consumes about 7 to 45 LSU per run and runs every six hours unless an admin disables it.
• Fleet and Sandboxes: Plus includes 37 LSU of Fleet usage and 8 LSU of Sandbox usage per month, then bills per LSU.
• Tuned Evaluators: 0.015 LSU per successful evaluation run, in public beta for Plus and Cloud Enterprise in the US.

Set usage limits under Settings, then Billing and Usage, so LangSmith converts a spend ceiling into trace caps, per the billing docs. Leaving Engine enabled on a busy project is the easiest way to get a surprise invoice.

Langfuse Pricing in 2026

Langfuse Cloud has four plans. Hobby is free with 50k units, two users, and 30 days of data. Core is $29 with 100k units and 90 days, Pro is $199 with three years of data and SOC 2 reports, and Enterprise is $2,499 with audit logs, SCIM, and an SLA, per Langfuse pricing.

Usage above the included 100k units is graduated: $8 per 100k units up to 1M, $7 up to 10M, $6.50 up to 50M, and $6 beyond that. The rates are identical across Core, Pro, and Enterprise, so you upgrade for retention, rate limits, and compliance rather than cheaper volume.

Langfuse add-ons and discounts

• Teams add-on: $300 per month on Pro for enterprise SSO, SSO enforcement, project-level RBAC, and a dedicated Slack channel.
• Startups: 50% off for the first year.
• Open source projects: $300 in monthly credits for the first year.
• Non-profits: $199 in monthly credits.

Self-hosted Langfuse cost

The MIT-licensed core is free to self-host with tracing, prompts, and evals included, per the self-hosting docs. Your real cost is infrastructure: Langfuse web and worker containers plus Postgres, ClickHouse, Redis, and S3-compatible storage. Enterprise features such as SCIM and audit logs need a paid license key, listed on the self-hosted pricing page.

Braintrust Pricing in 2026

Braintrust Starter is free with 1 GB of processed data, 10k scores, $10 of model credits, and 14-day retention. Pro is $249 per month with 5 GB, 50k scores, $100 of model credits, 30-day retention, custom charts, and RBAC, per Braintrust pricing.

Overage on Pro is $3 per GB and $1.50 per 1,000 scores, against $4 and $2.50 on Starter. Every Braintrust plan includes unlimited users, projects, datasets, and experiments, so team size never touches the bill.

Braintrust retention and model credits

Pro can keep data up to 180 days at $0.50 per GB per month beyond the included 30 days. Model credits pay for built-in AI features such as Topics and the Loop agent, and they also cover built-in models you call without your own API key; after the credit runs out you pay token rates.

Monthly Cost Breakdown: Three Real Workloads

We modeled one realistic agent run and priced it at three volumes. The assumptions are deliberately simple so you can swap in your own numbers.

• One run is one trace with eight observations (LLM calls and tool calls).
• Scoring: 10% of runs get one LLM-as-judge score.
• Payload: about 25 KB of processed data per run on Braintrust.
• Seats: 3, 5, and 10 people for the three workloads. Only LangSmith charges for them.
• Retention: each vendor’s default for the plan, which is not equal across vendors.

Workload	Langfuse Core	LangSmith Plus	Braintrust Pro	Braintrust Starter
Pilot: 20k runs, 3 seats	$35.56 (182k units)	$167 (3 seats + 10k traces)	$249 (0.5 GB, 2k scores)	$0
Growth: 100k runs, 5 seats	$93.80 (910k units)	$645 (5 seats + 90k traces)	$249 (2.5 GB, 10k scores)	$6
Scale: 1M runs, 10 seats	$668 (9.1M units)	$5,340 (10 seats + 990k traces)	$384 (25 GB, 100k scores)	$321

How we calculated each line

# Langfuse Core (graduated)
units = runs * 9 + runs * 0.10            # trace + 8 observations + scores
bill  = 29 + max(0, min(units, 1e6) - 1e5) / 1e5 * 8 + max(0, units - 1e6) / 1e5 * 7

# LangSmith Plus (base traces)
bill  = seats * 39 + max(0, runs - 10_000) * 0.005

# Braintrust Pro
bill  = 249 + max(0, gb - 5) * 3 + max(0, scores - 50_000) / 1000 * 1.50

At 1M runs, LangSmith Plus costs about 8x Langfuse Core and about 14x Braintrust Pro on list price. The gap comes almost entirely from the $5 per 1,000 trace rate, since seats add only $390 of the $5,340.

Hidden Costs That Change the Ranking

List prices only hold if your workload matches the model. These five variables move the ranking most often.

• Span depth: doubling observations per run roughly doubles the Langfuse bill and leaves LangSmith flat. At 30 observations per run and 1M runs, Langfuse Core passes $2,000.
• Payload size: long RAG contexts or multimodal inputs inflate Braintrust gigabytes. At 250 KB per run, the 1M-run Braintrust Pro bill rises to roughly $1,060.
• Retention parity: matching Langfuse Core’s 90 days is impossible on LangSmith Plus without extended traces, which add $2.50 per 1,000 traces, or about $2,500 at 1M runs.
• Score volume: scoring every run instead of 10% raises Braintrust Pro score charges from $75 to about $1,425 at 1M runs and adds 900k units on Langfuse.
• Platform extras: LangSmith Engine, Deployment, and Fleet bill separately in LSU, and Braintrust model credits run out on heavy Loop or Topics use.

Which Platform Is Cheapest for Your Team?

• Pick Langfuse for low to mid volume, many users, and long retention. Core at $29 is hard to beat below a few million units, and self-hosting removes the license bill. See Langfuse alternatives if the ClickHouse stack is the blocker.
• Pick LangSmith when you already run LangChain or LangGraph and want Deployment and Engine in one bill. Keep seats tight and cap traces. See LangSmith alternatives if trace cost is the dealbreaker.
• Pick Braintrust when eval gating in CI is the main job and run counts are high but payloads are small. The flat $249 covers most mid-size teams. Compare it with eval-first rivals in Braintrust vs Maxim vs Galileo.
• Stay on free tiers during a pilot. Braintrust Starter and Langfuse Hobby cover a 20k-run month at $0 if 14 or 30 days of history is enough.

How to Estimate Your Own Bill in 15 Minutes

1. Export one week of production runs and count traces, observations per trace, and average payload size.
2. Multiply to a 30-day month and add expected growth for the next quarter.
3. Decide how many runs you will score and how long you must keep data for audits.
4. Plug the numbers into the three formulas above, then add seats for LangSmith only.
5. Run the same week through each vendor’s free tier and compare the usage meter against your estimate before you sign.

Bottom Line on LangSmith, Langfuse, and Braintrust Pricing

Langfuse Core is the cheapest managed bill in most of our scenarios, Braintrust Pro wins at very high run counts with small payloads, and LangSmith is the premium option whose cost tracks trace count. Pay the LangSmith premium only when its suite, not its tracing, is the reason you are buying.

Related AgentStack Guides

This breakdown sits in Agent Observability under AgentOps. Start from the AgentOps tools map, then read Langfuse vs LangSmith vs Braintrust, best AI agent observability tools, LangSmith alternatives, Langfuse alternatives, AI agent evaluation tools, and open source LLM observability.

FAQ: LangSmith, Langfuse, and Braintrust Pricing