SSubrupt
Skip to content

Best LLM Observability Tools of 2026

Updated · 3 picks · live pricing · affiliate disclosure

Simpler analytics at twenty dollars monthly for thirty thousand events; the cheapest paid LLM observability tier.

BEST OVERALL6.4/10

Lunary

Simpler analytics at twenty dollars monthly for thirty thousand events; the cheapest paid LLM observability tier.

Free limited events; cancel-anytime
Try LunarySee full review

How it stacks up

  • Free limited events

    vs Helicone Pro $79

  • Team $20/mo + 30K events

    vs Langfuse Core $29

  • France-based EU residency

    Cheapest observability pick

#2
Helicone5.8/10

From $79/mo

View
#3
Langfuse5.8/10

From $29/mo

View

All picks at a glance

#PickBest forStartingScore
1LunaryBest cheap observability, $20/mo for 30K events$20.00/mo6.4/10
2HeliconeBest observability with proxy, one-line drop-in plus HQL$79.00/mo5.8/10
3LangfuseBest OSS observability self-host, MIT with traces plus evals$29.00/mo5.8/10

Quick pick by use case

If you only have thirty seconds, find your situation below and skip to that pick.

If I want gateway routing and observability in one productHeliconeHelicone Pro at seventy-nine dollars monthly with one-line drop-in proxy and HQL query language for analytics.If I want MIT-licensed observability with self-host postureLangfuseLangfuse Hobby free 50K observations a month; Core at twenty-nine dollars monthly upgrade. MIT OSS self-host free.If I want the cheapest paid LLM observability tierLunaryLunary Team at twenty dollars monthly for 30K events. France-based with EU data residency by default.

Compare all 3 picks

Top spec
#1Lunary6.4/10$20.00/mo$240.00/yr—Free limited events
#2Helicone5.8/10$79.00/mo$948.00/yr$708/yr moreHobby free 10K req/mo
#3Langfuse5.8/10$199.00/mo$2,388.00/yr$2,148/yr moreOSS MIT free self-host
#1

Lunary

6.4/10

Best cheap observability, $20/mo for 30K events

Try LunarySee Lunary alternatives

Simpler analytics at twenty dollars monthly for thirty thousand events; the cheapest paid LLM observability tier.

PlanMonthlyAnnualWhat you get
FreeFree—Free up to 10K events a month with the open-source Apache 2 core and self-hosted option
Team$20.00/mo$240.00/yr$20 a month for 30K events with prompt registry, analytics, and Slack alerts

Lunary is the cheapest LLM observability pick and the right call for teams that want straightforward analytics without paying for full feature breadth. Founded 2023 in France. The wedge for budget-conscious teams: Team tier at twenty dollars monthly covers thirty thousand events with logs, cost tracking, and basic prompt management; the cheapest paid tier in the LLM observability category. Free tier ships limited monthly events for evaluation.

Free covers limited monthly events with logs, cost tracking, and basic dashboards. Team is the upgrade tier at twenty dollars monthly with thirty thousand events plus collaboration features. Pro adds higher event limits and advanced analytics at the next upgrade tier. Enterprise covers custom event limits, SSO, and dedicated support. Most budget-conscious teams stay on Team indefinitely; Pro is the upgrade when event volume grows past thirty thousand monthly.

The trade-off versus Helicone is feature breadth; Lunary ships simpler analytics without HQL query language or per-property cost tracking. The trade-off versus Langfuse is OSS posture; Lunary is closed-source SaaS where Langfuse is MIT OSS with self-host. For budget-conscious teams who want functional observability without paying seventy-nine dollars or absorbing self-host operational tax, Lunary is the differentiated pick.

Pros

  • Team at twenty dollars monthly is the cheapest paid LLM observability tier
  • Thirty thousand events on Team covers small-team production traffic
  • Free tier with limited events for evaluation; no credit card required
  • France-based with EU data residency by default
  • Simpler dashboards than Helicone or Langfuse for low-complexity workflows

Cons

  • Simpler analytics versus Helicone HQL query language and Langfuse evals
  • Closed-source SaaS where Langfuse offers MIT OSS self-host
Free limited eventsTeam $20/mo + 30K eventsFrance-based EU residencyFree limited events; cancel-anytime

Best for: Budget-conscious teams who want functional LLM observability without paying for HQL or evals features and accept simpler analytics as the trade-off.

Routing
8
Latency
8
DX
8
Value
10
Support
7
Try Lunary
#2

Helicone

5.8/10$708/yr more

Best observability with proxy, one-line drop-in plus HQL

Try HeliconeSee Helicone alternatives

One-line drop-in proxy with cost analytics per session/user/property and HQL query language; YC W23.

PlanMonthlyAnnualWhat you get
HobbyFree—Free 10K requests a month with 1 GB storage, single seat, and one organization
Pro$79.00/mo$948.00/yr$79 a month with unlimited seats, alerts, reports, and the HQL query language; the realistic SMB paid entry
Team$799.00/mo$9,588.00/yr$799 a month with 5 organizations, SOC-2 + HIPAA compliance, and dedicated Slack support
Enterprise$2,000.00/mo$24,000.00/yrCustom contract with custom MSA, SAML SSO, on-prem deployment, and bulk cloud discounts

Helicone is the observability-first pick that combines gateway proxying with observability in one product. Founded 2023 (Y Combinator W23) in San Francisco. The wedge: one-line drop-in proxy that logs requests, tracks cost per session, user, and property, and ships an HQL query language for analytics. Set the base URL once and every OpenAI SDK call gets logged with full request and response, latency, and token cost.

Hobby is free with 10,000 requests a month, 1 GB storage, and a single seat. Pro is the upgrade tier at seventy-nine dollars monthly with unlimited seats, alerts, reports, and HQL. Team at the next upgrade tier covers 5 organizations, SOC-2, HIPAA compliance, and dedicated Slack support. Enterprise covers custom MSA, SAML SSO, on-prem deployment, and bulk discounts.

The trade-off versus Langfuse is OSS posture; Helicone Cloud is closed-source where Langfuse is MIT OSS with self-host. The trade-off versus Lunary is price; Helicone Pro at seventy-nine dollars monthly is more expensive than Lunary Team at twenty dollars. For teams that want observability and gateway in one product with cost tracking per session, Helicone is the right call.

Pros

  • One-line drop-in proxy: set base URL once, every OpenAI SDK call gets logged
  • Cost analytics per session, user, and property with HQL query language
  • Hobby free covers 10,000 requests a month with 1 GB storage
  • Pro at seventy-nine dollars monthly covers unlimited seats with alerts and reports
  • Y Combinator W23; combines gateway routing with observability

Cons

  • Pro repriced from twenty to seventy-nine dollars monthly in 2025 (a 295 percent increase)
  • Cloud is closed-source where Langfuse offers MIT OSS self-host
Hobby free 10K req/moPro $79/mo + unlimited seatsTeam $799/mo + SOC-2Hobby free 10K req/mo; cancel anytime

Best for: Teams that want observability and gateway in one product with cost tracking per session, user, and property and accept closed-source Cloud as the trade-off.

Routing
9
Latency
8
DX
9
Value
7
Support
8
Try Helicone
#3

Langfuse

5.8/10$2,148/yr more

Best OSS observability self-host, MIT with traces plus evals

Try LangfuseSee Langfuse alternatives

MIT-licensed LLM observability with traces, evals, and prompt management; self-host option; YC W23 in Berlin.

PlanMonthlyAnnualWhat you get
HobbyFree—Free 50K observations a month with all platform features (limits apply), 30-day data retention, and 2 users
Core$29.00/mo$348.00/yr$29 a month for 100K observations with unlimited users, 90-day retention, and in-app support
Pro$199.00/mo$2,388.00/yr$199 a month with unlimited history, 3-year data access, SOC 2, ISO 27001, HIPAA, and 20K req/min throughput
Enterprise$2,499.00/mo$29,988.00/yr$2,499 a month with audit logs, SCIM API, custom rate limits, dedicated support engineer, and uptime SLAs

Langfuse is the MIT-licensed OSS observability pick and the most-popular OSS LLM observability project. Founded 2023 (Y Combinator W23) in Berlin Germany. Around 6,000 GitHub stars by 2026. The wedge: MIT-licensed observability with traces, evals, prompt management, and analytics on a project that you can run on your own infrastructure for unlimited observation volume.

Hobby is free with 50,000 observations a month, all platform features (with limits), 30-day data retention, and 2 users. Core is the upgrade tier at twenty-nine dollars monthly with 100,000 observations and unlimited users plus 90-day retention. Pro at the next upgrade adds 3-year retention, SOC 2, ISO 27001, HIPAA, and 20K req/min throughput. Enterprise covers audit logs, SCIM API, dedicated support, and SLAs.

The trade-off versus Helicone is gateway routing; Langfuse is observability-only where Helicone bundles gateway routing. The trade-off versus Lunary is price floor; Langfuse Core at twenty-nine dollars monthly is slightly more expensive than Lunary Team at twenty dollars. For OSS readers who want self-hostable observability with unlimited volume, Langfuse is the right call.

Pros

  • MIT permissive license; all commercial use allowed at any revenue level
  • Self-host option for unlimited observation volume on your own infrastructure
  • Hobby free covers 50,000 observations a month with all features and 2 users
  • Around 6,000 GitHub stars by 2026; the most-popular OSS LLM observability
  • Berlin-based with default EU data residency on Cloud

Cons

  • Observability-only; gateway routing requires a separate proxy layer (LiteLLM)
  • Self-host requires Postgres plus Node.js operational maturity
OSS MIT free self-hostHobby free 50K obs/moCore $29/mo upgradeHobby free 50K obs/mo; MIT OSS self-host free forever

Best for: OSS readers who want MIT-licensed observability they can self-host with unlimited observation volume; pair with LiteLLM for full OSS stack.

Routing
10
Latency
8
DX
8
Value
9
Support
7
Try Langfuse

How we picked

Each pick gets a transparent composite score from price, features, free-tier availability, and editor fit. Pricing flows from our live database, so when a vendor changes prices the score updates here too.

We weight price at 40 percent, features at 30, free tier at 15, fit at 15. Lunary leads on price at twenty dollars monthly; Langfuse leads on MIT OSS posture; Helicone leads on combining gateway routing with observability. See the parent /best/llm-gateways guide for pure-routing picks excluded from this lens.

We don't claim "30,000 hours of testing." Our methodology is the formula above plus the editor's published verdict for each pick. Verifiable, auditable, and updated when the underlying data changes.

Why trust Subrupt

We're a subscription tracker first, a buying guide second. Every claim on this page is something you can check.

By use case

Best observability with proxy

Helicone

Read the full review →

Try Helicone

Best OSS observability self-host

Langfuse

Read the full review →

Try Langfuse

Cheapest observability

Lunary

Read the full review →

Try Lunary

How to choose your LLM Observability Tool

Observability layer vs gateway routing layer

LLM observability sits at a different layer from gateway routing. Gateway routing handles which provider serves a given request (OpenRouter, Cloudflare, Vercel, LiteLLM, Portkey). Observability handles logging, tracing, evaluating, and cost-tracking the requests after they have been routed (Helicone, Langfuse, Lunary). The two layers can stack: a team running LiteLLM as the proxy layer plus Langfuse as the observability layer gets full OSS coverage at zero per-request fees. Helicone uniquely combines both layers in one product: it is a one-line proxy that also ships the observability dashboard. The decision pivots on whether you want one combined tool (Helicone) or two specialized tools (LiteLLM plus Langfuse, or any proxy plus Lunary).

Trace depth: spans, sessions, and prompt management

Observability trace depth varies sharply. Helicone logs each request with full request and response payloads, latency, token count, and cost; HQL query language lets teams analyze across sessions, users, and properties. Langfuse ships hierarchical traces (spans nest inside sessions inside traces), evaluations against datasets, prompt versioning with rollout, and analytics. Lunary ships simpler logs with cost tracking but no HQL or hierarchical span model. For teams running agentic workflows with multi-step tool calls, Langfuse's hierarchical trace model is load-bearing. For teams running simple chat completions, Helicone or Lunary cover the use case more cheaply.

Eval support: dataset-driven evaluation pipelines

Eval pipelines run model outputs against datasets to measure quality regression over time. Langfuse ships native eval support with dataset management, automated eval runs, and trend visualization on Hobby and above. Helicone ships basic eval support on Pro at seventy-nine dollars monthly. Lunary ships limited eval features on Team and Pro tiers. For AI teams whose primary workflow is dataset-driven evaluation (like Vellum or Braintrust at higher prices), neither Helicone nor Lunary is the right call; Langfuse is the closest catalog fit. For teams where eval is an occasional check rather than a primary workflow, all three picks cover the basics.

Self-host vs managed Cloud trade-offs

Langfuse is the only catalog observability pick that ships full self-host as MIT OSS. Self-host runs on Postgres plus the Langfuse Node.js application; the operational tax is database backups, migrations, and observability ingestion scaling. For teams without DevOps capacity, Langfuse Cloud Hobby covers fifty thousand observations a month free and Core covers one hundred thousand observations at twenty-nine dollars monthly. Helicone offers self-host on Enterprise tier only; Hobby and Pro are managed Cloud. Lunary is managed Cloud only. For OSS-aligned teams that need self-host posture for compliance or cost-at-scale, Langfuse is the only fit. For teams comfortable with managed Cloud, Helicone Pro or Lunary Team trade managed convenience for the per-event price.

When to consider full LLM gateway picks (cross-link to parent)

Pure observability picks cover the logging and tracing layer but do not handle multi-provider routing, fallbacks, or guardrails. Some teams need both layers in one product. Helicone is the only catalog pick that combines gateway routing with observability; Pro at seventy-nine dollars covers SMB scale. For production teams that want full gateway features (caching, prompt management, RBAC) plus observability, Portkey Production at forty-nine dollars covers the broader feature set on Cloud (or Apache 2.0 self-host since March 2026). At those decision points, see [our /best/llm-gateways guide](/best/llm-gateways) for the broader lineup including pure-routing picks like OpenRouter and Cloudflare AI Gateway plus full-gateway picks like Portkey.

Frequently asked questions

Why is Helicone ranked first over cheaper Lunary?

Helicone wins on combined gateway-and-observability defensibility because the one-line drop-in proxy plus HQL query language plus cost analytics per session is the most polished observability profile in catalog LLM observability. Lunary wins on price specifically. Teams that want gateway routing and observability in one tool pick Helicone. Budget-conscious teams that want functional observability without HQL pick Lunary.

Does Subrupt earn a commission from these observability picks?

On a few. We disclose this on every /best page. Self-host OSS (Langfuse MIT) has no affiliate path because there is no transaction. Cloud upgrades on Helicone, Langfuse, and Lunary have paid plans where we earn commission only on conversion. The composite ranking weights price at 40 percent, features at 30, free tier at 15, fit at 15; none tuned by affiliate rate.

Should I run Helicone or Langfuse plus a proxy?

Helicone bundles gateway proxying with observability in one product; set the base URL once and every OpenAI SDK call gets logged plus routed. Langfuse plus a proxy splits the layers; you run LiteLLM (proxy) plus Langfuse (observability) as two specialized OSS tools. Helicone is friction-free for teams that want one tool covering both layers. Langfuse plus LiteLLM is friction-free for teams that want self-host posture and accept running two services.

Is Lunary really cheaper than Langfuse?

Slightly. Lunary Team is twenty dollars monthly for thirty thousand events; Langfuse Core is twenty-nine dollars monthly for one hundred thousand observations. Per-event, Lunary works out to around sixty-six cents per thousand; Langfuse to around twenty-nine cents per thousand. For low-volume workloads Lunary is cheaper absolute; for higher-volume workloads Langfuse wins per event. Self-hosted Langfuse OSS is dollar-zero on licensing.

How much does self-hosting Langfuse actually cost?

A small Langfuse deployment runs on a single VPS plus managed Postgres at around fifteen to thirty dollars monthly. The operational tax is database backups, schema migrations, and observability ingestion scaling. For staging and prod, double the cost. Above that, you absorb patching, monitoring, and on-call. Cost is comparable to Langfuse Cloud Core at twenty-nine dollars monthly; the trade-off is operational labor versus managed.

What about Langsmith and Arize?

Langsmith is the LangChain-native observability tool at thirty-nine dollars per user monthly; tightly coupled with LangChain agents but limited beyond that ecosystem. Arize is the enterprise ML observability platform with strong eval features at higher price points. Both are out of our catalog; from our catalog Helicone, Langfuse, and Lunary are the LLM observability picks. Teams already deep in LangChain should evaluate Langsmith alongside Langfuse.

Can I migrate observability data from Helicone to Langfuse?

Partially. Both export request logs as JSON via dashboard or API. Helicone HQL queries do not translate directly to Langfuse SQL queries; analytics dashboards need rebuilding. Trace structures differ; Helicone is request-based where Langfuse is hierarchical span-based. Plan one to two weeks for a non-trivial migration. New telemetry routes to the new tool from migration day; historical data backfill is one-shot.

EU data residency: which observability picks store data in the EU?

Langfuse is Berlin-based with default EU residency on Cloud; self-host on EU infrastructure gives full control. Lunary is France-based with EU residency by default. Helicone Cloud is US-default; Enterprise covers EU on-prem. For EU-resident observability use, Langfuse and Lunary are the cleanest defaults; Helicone requires Enterprise tier or self-host.

How often is this guide updated?

We re-review pricing and feature changes annually at minimum, with mid-year refreshes when major vendor announcements happen. Helicone Pro repriced from twenty to seventy-nine dollars monthly in 2025. Langfuse Core launched at twenty-nine dollars monthly in 2024. Lunary Team pricing has been stable. The lastReviewed date reflects the most recent editorial pass.

Subrupt Editorial

The team behind subrupt.com. We track subscriptions, surface cheaper alternatives, and publish buying guides where the score formula is on the page so you can recompute it yourself. We do not claim 30,000 hours of testing. What we claim is live pricing from our database, a transparent composite score, and honest savings math against a category baseline.

Last reviewed

Citations

Affiliate disclosure: Subrupt earns a commission when you switch to a service through our recommendation links. This never changes the price you pay. We only recommend services where there's a real cost or feature advantage for you, and our picks are based on the data on this page, not on which programs pay the most.

Related buying guides

Buying guide

Best Threat Intelligence Platforms of 2026

Read guide

Buying guide

Best VPNs of 2026

Read guide

Buying guide

Best Free VPNs of 2026

Read guide

Track your subscriptions on Subrupt

Add the LLM Observability Tool you pay for and see how much you'd save by switching.

Open dashboard

More buying guides

Independent rankings for the subscriptions worth paying for.

See all guides