GitHub's Fusion routes each task across models billed at their own rates, with a 30% launch discount. Plan prices, credit pools, the Claude Code comparison and new Automations caps.
GitHub made Copilot Fusion generally available on 28 October 2026. Instead of picking a model per task, developers choose Fusion and Copilot routes and orchestrates models for them. It runs on every Copilot plan in VS Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs, with more clients "coming soon". On the same day GitHub added credit caps to Copilot Automations.
How Fusion works
Fusion has three effort levels: Standard (everyday coding), High (harder tasks) and Max (the most demanding work). Behind them are three workflows. Single has one model solve the task. Cascade has an efficient model draft an answer and a quality check accepts it or escalates to a stronger model. Critique has one model draft, another review, and the first revise. It folds the HydraFusion research preview and Copilot auto model selection together, and users of those are moved to the matching Fusion level automatically.
GitHub does not say which models Fusion routes to, and it publishes no per-model rates on these pages.
Billing and the 30% discount
You are billed for each model Fusion uses at that model's standard rate. Paid-plan users get 30% off model costs when using Fusion "for a limited time"; no end date is given. Copilot now meters in AI credits, where one credit is $0.01. Code completions and next edit suggestions do not use credits; agent mode, code review, the cloud agent, CLI and the app do. For how seats and credits interact on the enterprise side, see our earlier piece, GitHub Enterprise now includes Copilot at $0.
How a Copilot plan's credits flow through Fusion's levels and workflows, with each model billed at its own rate less the 30% launch discount.
Plan numbers
| Copilot plan | Price | Credits per month |
|---|---|---|
| Pro | $10 | $15 ($10 base + $5 flex) |
| Pro+ | $39 | $70 ($39 base + $31 flex) |
| Max | $100 | $200 ($100 base + $100 flex) |
Max is listed at "2.9x+ included usage" versus Pro+. Credit purchases are available on Pro, Pro+ and Max.
Against that, Claude Pro costs $20 a month ($17 on annual billing, $200 up front) and Claude Max costs $100 a month with "5x Pro usage" every five hours and week; Claude Code is included in both. Anthropic does not publish a dollar figure for those allowances, so the two are not directly comparable: Copilot gives a stated dollar pool, Claude gives relative rate limits.
On the API, Claude Opus 5.5 is $4 per million input tokens and $20 per million output tokens, with cache hits at $0.20. That is below Opus 5 at $5 and $25.
The cost-per-task claim
GitHub reports that on the Artificial Analysis Coding Agent Index, Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max. It says Fusion used about half the average API cost per task, about half the tokens and about 40% less time. This is GitHub's own citation of the benchmark and we have not independently verified it. Both vendors list a $100 a month tier.
Copilot Automations credit caps
Automations now accept optional AI-credit limits per run, per day, per week and per month, and the limits can be combined. A run stops at its per-run limit; once a daily, weekly or monthly cap is hit, new runs do not start until the period resets. Automations without limits behave as before, so uncapped scheduled runs can still drain a plan's credits. GitHub gives no default values.
The same update adds an API trigger (examples: Jira, Sentry, PagerDuty), 24 new GitHub triggers across issues, pull requests, discussions and security alerts, chaining through "Automation run succeeded" and "failed" triggers, 20 new GitHub tools and support for custom MCP server tools. Automations are available on Pro, Pro+, Max, Business and Enterprise, in any repository where you have write access.
What to do
- Set per-run and daily caps on every Automation before enabling the new triggers, especially security-alert and PR-event triggers that can fire in bursts.
- Test Fusion on your own tasks before trusting the half-the-cost claim; track credits per merged task.
- Enterprise admins: review which models are enabled for Fusion, since it only uses models allowed by policy, and consider Fusion-only model restrictions.
- Treat the 30% discount as temporary when budgeting.
Sources
- https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion
- https://github.blog/changelog/2026-10-28-new-tools-triggers-and-cost-controls-for-copilot-automations
- https://github.com/features/copilot/plans
- https://claude.com/pricing
- https://platform.claude.com/docs/en/about-claude/pricing