Langfuse v4 went live on Langfuse Cloud and generally available for self-hosted deployments on 17 August. The change is that every LLM call, tool execution and agent step becomes directly queryable and evaluable, rather than reachable only through a trace view: initial table loads over large datasets drop from seconds to milliseconds, and dashboards over longer time ranges load at least ten times faster in large projects.
There is a date attached, which is the part teams will feel. Langfuse Cloud becomes v4-only on 16 November 2026; most projects need no migration, but a Migration Assistant surfaces per-project actions that must be completed before that date, and self-hosted operators are told to plan the upgrade and review infrastructure requirements first.
Making the agent step the queryable unit is what turns observability into a diagnostic rather than a record, which is the transition described in evals moving from leaderboards to operating diagnostics and part of the control plane in agent operations converging on connectors, memory and observability. The agents hub tracks that layer.
The performance figures are the vendor's own and stated as ranges for large projects rather than a published benchmark. The condition to watch is whether step-level querying makes evaluator results comparable across versions, because faster dashboards over the same ambiguous step boundaries would improve latency without improving diagnosis.