Replit launches effort-based pricing - per-checkpoint $0.25 model replaced
On June 18, 2025 Replit replaced its $0.25-per-checkpoint Agent pricing with effort-based pricing that bundles complex tasks into single variable-cost checkpoints. Rolled out to new users immediately, forced on existing Core and Teams subscribers July 1.
Variable per checkpoint - effort-metered - opaque
Replacing a deterministic per-checkpoint price with an opaque effort metric removes user ability to predict task cost; subsequent Agent 3 overruns validated the concern.
Opaque metering shifts the cost-prediction burden from vendor to user without compensating disclosure; the structural change was the prerequisite for later cost surprises.
Effort-based billing can be legitimate, but launching it without per-task cost preview tooling left users navigating bills retrospectively.
Replit responded on day 376; their statement is attached below.
Prior to the change, every Replit Agent checkpoint cost a fixed $0.25 regardless of complexity.
Prior to the change, every Replit Agent checkpoint cost a fixed $0.25 regardless of complexity. The new model claims to charge less for simple edits and more for complex multi-step work. Replit's own announcement noted: 'Simple changes still result in a single checkpoint - typically costing less than $0.25. Larger or more complex tasks are no longer split across multiple $0.25 checkpoints.' Critics quickly pointed out that what counts as 'effort' is undocumented, leaving users with no way to predict task cost before submission. This change is the structural foundation for the September 2025 Agent 3 cost-overrun incident and the November 2025 forced-migration receipt already catalogued.
Open-source Claude Code / Cursor replacement that lives in the terminal. Pick when you want zero vendor lock-in and any LLM backend.
Permissive (MIT) open-weight model targeting ChatGPT/Claude-class chat and agentic coding without API lock-in - the strongest open option for a self-hosted stack.
ChatGPT's Code Interpreter, but local and against any model.
Near-frontier open-weight model to back a self-hosted ChatGPT/Claude replacement or an agentic coding stack.