GitHub Copilot moved from flat-fee ($10-19 per developer per month) to usage-based billing on June 1, 2026, and the first full monthly cycle closed June 30. Enterprise buyers are now reviewing their first real bills — and initial reports suggest average charges are 25-40% above the equivalent flat-fee bundle, particularly for teams making heavy use of Copilot Workspace agent features.
The pricing structure charges separately for completions, chat interactions, agent invocations and code review sessions. Simple autocomplete is cheap; long-horizon agent runs that make repeated model calls are far more expensive. Enterprises with 200+ developers are the most exposed because heavy agent adopters can single-handedly balloon monthly spend.
“Enterprises with 200+ developers are the most exposed because heavy agent adopters can single-handedly balloon monthly spend.”
Context for the shift: GitHub's own Copilot cost of goods sold has ballooned with GPT-5 and GPT-5.5 inference, making flat-fee pricing unsustainable — a pattern Cursor, Sourcegraph and Anthropic Code all mirrored across 2026. Usage-based is now the industry standard for developer AI.
Competitive landscape: Cursor sells $20/mo Pro and $40/mo Business tiers with generous quotas that reset monthly; Windsurf offers a hybrid; Sourcegraph Cody offers per-seat with usage overages. Each is trying to lean into 'predictable pricing' as marketing wedge against Copilot's exposed pricing volatility.
What to watch: whether GitHub responds with usage caps or budget alerts (announced roadmap items), whether enterprises consolidate to a single vendor to simplify governance, and whether spend-management tools like Ramp add native Copilot integrations for line-item tracking.