Helicone vs Langfuse: Cost Tracking, Enforcement, and Who Maintains Them

Where each tool sits in the request path

Helicone's main integration is a base URL change. Requests go to Helicone's AI Gateway, which forwards them to the provider and logs the result. Because every request passes through it, its plans include gateway features: caching, rate limits, and automatic fallbacks are listed on the Helicone pricing page as of 2026-10-08.

Langfuse integrates through Python and JavaScript SDKs, OpenTelemetry, framework integrations, or proxy-based logging through LiteLLM, according to its pricing page feature table. It records traces, sessions, and observations, then calculates cost from ingested usage or from its own model price definitions. The two are not exclusive. Langfuse publishes an integration guide for sending traffic through Helicone's gateway while tracing in Langfuse.

Cost tracking versus cost enforcement

Tracking tells you what you spent. Enforcement stops the next request. Helicone documents a cost-based limit through the Helicone-RateLimit-Policy header: setting the unit to cents turns the quota into a spend cap, scoped globally, per user, or per custom property such as organization. A policy of 500;w=3600;u=cents;s=user caps each user at $5.00 per hour. Requests over the limit receive a 429 error, and the minimum window is 60 seconds. Token-based limits and multiple policies per request are listed as coming soon.

Langfuse's cost tracking is detailed but after the fact. Its docs say cost is recorded only on generation and embedding observations, that model definition changes apply only to new generations, and that reasoning models need ingested usage because Langfuse cannot infer their token counts. For a hard stop on spend, Langfuse needs a gateway or provider limit in front of it.

• Use Helicone, or another gateway, when you need a per-user or per-tenant dollar cap enforced before the provider bills you.
• Use Langfuse when you need to see which step of a multi-step agent run spent the money.
• Use both when you need enforcement and span-level attribution, since Langfuse documents the Helicone gateway integration.

Pricing units are not comparable

Helicone counts requests and storage. Its free Hobby plan includes 10,000 requests, 1 GB of storage, one seat, and 7 days of retention. Pro is $79 per month with unlimited seats and 1 month of retention, and Team is $799 per month with 3 months of retention and SOC 2 and HIPAA compliance. Paid plans add usage-based charges beyond the free 10,000 requests and 1 GB.

Langfuse counts units, defined as traces plus observations plus scores. Hobby is free with 50,000 units and 30 days of history. Core is $29 per month with 100,000 units included, Pro is $199 per month with 3 years of history, and Enterprise is $2,499 per month. Extra units are graduated: $8 per 100,000 up to 1 million, $7 up to 10 million, $6.50 up to 50 million, and $6 beyond that. Self-hosted Langfuse is free under the MIT license.

The unit difference matters for agents. A Langfuse trace with five observations is six units, and its billable units page says scores created by LLM-as-a-judge evaluators count too. An application with 200,000 traces a month at five observations each is 1.2 million units, which costs $115 per month on Langfuse Core: the $29 base, $72 for units 100,000 to 1 million, and $14 for the last 200,000. Helicone counts each LLM request plus stored data instead, so its logging cost follows request volume and payload size rather than how many spans you record per run.

What the 2026 acquisitions changed

ClickHouse acquired Langfuse in January 2026. Langfuse's announcement says the roadmap stays the same, the product stays open source and self-hostable with no planned licensing changes, and the team joins ClickHouse to keep building it. Langfuse already used ClickHouse as its core data store since v3, and it shipped Langfuse v4 in August 2026.

Mintlify acquired Helicone on March 3, 2026. Helicone's announcement says its services remain live for the foreseeable future in maintenance mode, with security updates, new models, and bug and performance fixes continuing to ship. The announcement describes keeping the service running and does not describe new feature work. Teams choosing Helicone today should plan for the gateway they adopt now being close to the gateway they will have in a year, and keep the base URL swap reversible.