The free ride for heavy AI users is over. GitHub announced a hard pivot to usage based billing for Copilot, effective June 1, ditching the old model where a quick chat cost the same as an all day autonomous coding binge. The Microsoft owned company admitted it can no longer absorb the escalating inference costs, a problem that exploded with the rise of agentic assistants like Openclaw that burn through tokens like kindling. Under the new scheme, subscribers get a monthly allotment of AI Credits tied to their plan, with overage charged at per model API rates. That means GPT-5.5 output will hit you at $30 per million tokens, while simpler code completions remain credit free. GitHub is also slapping extra costs on code reviews via Actions minutes. This is a direct response to unsustainable cost spikes, with internal documents reportedly showing week over week Copilot costs nearly doubling since January. Users can preview their projected bills before the change takes effect.
The End of All You Can Eat AI
This is the second major signal in a week that the era of flat fee AI subscriptions is dying. Anthropic recently started charging its largest Claude Enterprise customers for full compute costs, not subsidized tokens, and briefly tested stripping resource hungry Claude Code from the $20 Pro plan. Their peak hour throttling (5 am to 11 am Pacific) is a transparent attempt to cap losses. When companies like Microsoft and Anthropic cry about sustainability, what they really mean is they cannot turn massive demand into actual profit. The VC funded race to hoard users by undercutting price is over. Now they need to extract real revenue from the heaviest users, the ones who treat AI like an infinite intern. Expect every major vendor from OpenAI to Google to follow suit with granular token metering. The free lunch has been cleared from the table.
Source: Arstechnica
