GitHub Copilot introduces usage-based billing on June 1, 2026
GitHub says Copilot will shift from flat-rate plans to usage-based billing starting June 1 as agentic features expand. The change makes token budgeting a first-order engineering constraint and adds more pressure on teams comparing Copilot with other coding agents.
TL;DR
- GitHub says Copilot will switch to a repost of GitHub's announcement on June 1, replacing flat-rate plans with usage-based billing as Copilot adds more agentic and advanced features.
- A screenshot of GitHub's multiplier table shows the price pressure is not evenly distributed: Claude Sonnet 4 stays at 1x, while Claude Sonnet 4.5 jumps to 6x and Claude Opus 4.7 jumps to 27x.
- Gergely Orosz's reaction frames the move as part of a broader retreat from subsidized AI subscriptions, and kilocode's thread points to similar pricing stress across GitHub, Cursor, and Windsurf.
The weirdly useful bit is the model table. GitHub's multiplier screenshot suggests Copilot is starting to price frontier models like expensive infrastructure instead of bundling them into a feel-good subscription. kilocode's thread adds that the shift is landing alongside other plan resets across coding tools, which makes this look less like a one-off pricing tweak and more like the new baseline.
June 1 billing switch
GitHub's announcement, surfaced in the reposted notice, says Copilot moves to usage-based billing on June 1 because the product now includes more agentic and advanced capabilities.
That wording matters more than the date. GitHub is explicitly tying pricing to higher-cost workflows, not just autocomplete.
Model multipliers
The most concrete detail in the evidence is the screenshot of GitHub's multiplier table. It lists current and new multipliers side by side:
- Claude Haiku 4.5: 0.33x to 0.33x
- Claude Sonnet 4: 1x to 1x
- Gemini 2.5 Pro: 1x to 1x
- Claude Sonnet 4.5: 1x to 6x
- Claude Sonnet 4.6: 1x to 9x
- Claude Opus 4.5: 3x to 15x
- Claude Opus 4.6: 3x to 27x
- Claude Opus 4.7: 3x to 27x
The table makes Copilot's new logic pretty plain: cheap models stay near baseline, while newer Anthropic models get repriced hard.
Flat-rate pressure across coding tools
The commentary around the announcement is broader than Copilot itself. Gergely Orosz calls it the beginning of the end for subsidized AI subscriptions, while kilocode's thread argues the same pressure was already visible in plan changes elsewhere.
kilocode's list is specific:
- GitHub paused Pro signups
- GitHub pulled Claude Code from the $20 tier
- Cursor marked its $60 plan as the recommended option over $20
- Windsurf moved from $5 to $20
That does not prove every coding agent will converge on the same model. It does show multiple vendors trying to stop hiding expensive model usage behind flat monthly pricing.