Devin's ACU consumption per task silently rebalanced upward
Cognition Labs adjusted what counts as one Agent Compute Unit (ACU), increasing actual consumption per task by an estimated 30-50%. Devin's $500/mo plan with 10 ACU/mo now buys meaningfully less work.
$500 / mo · 30-50% more ACU per same task
Devin did not respond inside the 30-day window. This receipt now reads as uncontested in the public record.
Devin charges by ACU, an opaque unit that bundles compute time, model calls, and tooling overhead.
Devin charges by ACU, an opaque unit that bundles compute time, model calls, and tooling overhead. The rebalance was attributed to 'more accurate accounting of background processes' but no per-task cost-before/after data was published.
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.