How does Langfuse compare with the alternatives?
Pick Langfuse when trace data must stay on your servers without a contract: its MIT-licensed core self-hosts free, and Cloud bills units of traces, observations, and scores. Pick LangSmith, billed per seat and per trace, for one vendor that traces and hosts LangGraph agents; it self-hosts only on Enterprise.
bbv left Arize Phoenix for Langfuse over Phoenix's license, and Bayes Impact moved to Phoenix for a lighter stack. Pick Langfuse to self-host free under the MIT license with prompt management included. Pick Arize AX when the same team also monitors classification or ranking models for drift and performance.
CREATE SOMETHING retired Braintrust for Langfuse in July 2026. Pick Langfuse to run the whole platform yourself; at 200,000 traces a month GILD's calculation from published rates puts its Pro plan at US$369 against US$474 on Braintrust Pro. Pick Braintrust for a managed eval product when scores per trace stay few.
Both capture Claude Code sessions through its hooks. Pick Langfuse to trace and evaluate an LLM application you build, metered in units. Pick kcap to keep every coding agent's sessions as team memory priced per seat.
