From Flat Subscriptions to Token-Based Billing
GitHub Copilot’s new token-based billing model is a usage-based pricing system where developers pay according to AI tokens consumed, measured in credits that reflect the cost of prompts, responses, and cached context instead of fixed request quotas. Copilot previously mixed a standard monthly subscription with “premium request” units that covered heavier features such as advanced chat and long-running agent sessions. Those premium requests were detached from real infrastructure costs, and GitHub often shouldered the bill when users ran intensive workloads for hours. With that model retired, every Copilot plan now includes a pool of GitHub AI Credits. These credits are spent based on token usage for each supported model, so long chats and large context windows now show up as direct cost. Code completions and next edit suggestions remain included in paid plans and do not consume credits.
How the New AI Credits System Works
Under the new AI credits system, each plan’s price stays the same but translates into a mix of base and “flex” credits that track token usage. One AI credit equals one cent of usage, and credits are consumed by input tokens, output tokens, and eligible cached data, at rates that depend on the chosen model. The Pro plan at USD 10 (approx. RM46) per month includes USD 15 (approx. RM69) worth of credits, while Pro+ at USD 39 (approx. RM179) includes USD 70 (approx. RM322). According to The New Stack, “The flex allotment is a variable part of your included usage; it is designed to adapt as the economics of AI evolve.” Flex credits were added after users warned that base amounts might not cover heavy agent sessions, giving GitHub room to adjust as model pricing and efficiency change.
Copilot Max and Cost Control for Heavy Users
For developers who hammer Copilot with long, agentic sessions, GitHub has introduced Copilot Max as a higher ceiling. Priced at USD 100 (approx. RM460) per month, Max includes USD 200 (approx. RM920) in total monthly credits, split into USD 100 (approx. RM460) in base credits plus a USD 100 (approx. RM460) flex allotment. That puts it well above Pro+ and aims to support sustained, high-volume workloads without constant limit hits. Plan prices themselves remain unchanged; what changes is how far your credits stretch, depending on model choice and token volume. Smaller models such as GPT-5.4 nano can generate one million output tokens for about USD 1.25 (approx. RM6), while the same volume from a frontier-class GPT-5.5 model is around USD 30 (approx. RM138), so Max users still need to pay attention to which options they select.
Enterprise Budgets, Pooled Credits, and Admin Controls
Business and enterprise Copilot tiers keep their per-seat prices at USD 19 (approx. RM87) and USD 39 (approx. RM179) per user per month, with matching credit allotments for each seat but no flex portion. Instead of separate buckets for every developer, credits are pooled at the organization level, so power users can draw more while lighter users offset their consumption. Through August, Business seats receive USD 30 (approx. RM138) in promotional credits and Enterprise seats receive USD 70 (approx. RM322). The new budget control system adds several layers of governance, including user-level budgets, individual overrides, cost center budgets, and an enterprise-wide cap to prevent runaway spend. Alongside billing changes, Copilot code review now consumes both AI credits and GitHub Actions minutes, which are billed at the same per-minute rate as other Actions workflows, with admins able to set organization-wide runners.
What Developers Are Seeing: Faster Spend, Clearer Trade-Offs
The move to a metered, token-based billing system is exposing how expensive certain AI habits can be. Under the old premium request model, a quick chat and a multi-hour refactor both counted as single units, hiding the real cost of long sessions, wide context windows, and frontier models. With AI credits, developers are now watching usage meters climb quickly. TechSpot reports that a simple “build a Minesweeper game” prompt with Claude Haiku 4.5 consumed about 94 credits, and some users say they have burned through what felt like “months” of usage in a single day since the rollout. For many teams, this is pushing a change in behavior: switching to smaller models for routine tasks, trimming prompt size, and reserving the most powerful models for production-critical work where the extra spend is easier to justify.
