Our Take
Microsoft sold Copilot subscriptions as loss leaders and finally priced it to reality, but failed to warn users the subsidy was ending.
Why it matters
Developers who budgeted for flat-rate Copilot now face surprise overage costs under token counting. This affects individual contributors and teams still onboarded to the old model.
Do this week
Engineering leads: audit your Copilot spend for the past 30 days against your monthly budget and flag to finance before next billing cycle.
Token billing exhausted credits in hours, not weeks
GitHub migrated Copilot billing from fixed monthly subscriptions to token-based pricing in April 2026. Within the first day, users reported credit depletion rates far exceeding their historical consumption under the flat-rate model.
User 'rvs99' documented a single use case where minor edits to six files cost approximately $0.35 per line of actual code change. Another user, 'prhost', shared a dashboard screenshot showing 3,705 credits burned in one day from an initial allowance of 7,000 credits (company-reported). A third user, 'zoomp05', summarized the pattern: teams expected the transition to be cost-neutral or cheaper, only to discover their remaining budget depleted at multiples of their prior spending rate.
The shift follows months of speculation in GitHub's community discussions after Microsoft announced the change in April. Users had questioned whether token-based billing would cost more or less than subscriptions. The answer arrived immediately: significantly more for typical development workflows.
Subscriptions were never intended to scale profitably
Microsoft's original Copilot subscription offering functioned as a loss leader. The company absorbed the gap between the monthly subscription fee and the actual cost of LLM inference, training, maintenance, and infrastructure. That gap was always temporary.
Running LLMs at scale is expensive. Beyond inference, costs include model development, post-training, data center construction, hardware amortization, and energy. Subscription billing that allowed users to burn far more tokens than their fee covered was unsustainable by design. Microsoft's move to token-based pricing aligns with supplier costs, not with user expectations formed during the subsidized trial period.
The surprise among users reflects Microsoft's failure to telegraph the end of the subsidy. The company did not frame the transition as the expiration of promotional pricing. Instead, it moved to consumption-based billing without a clear communication that users had been getting a discount.
Audit your consumption and budget before the next cycle
Developers and teams should pull their Copilot usage logs for the past 30 days and calculate the token-based cost under the new pricing. Compare that to your current subscription spend. If the token burn rate from that first week holds steady, budget accordingly or evaluate alternatives.
If your team's Copilot usage is concentrated among a few power users, consider restricting token allocation per developer or rolling out adoption in phases rather than enabling all seats at once. If burn rates remain high after audit, weigh the trade-off between Copilot's productivity value and the cost of alternatives such as local fine-tuned models or competitors with different pricing models.
Communicate the cost change to finance and stakeholders now. Do not wait for the next billing surprise.