GitHub Copilot Moves to Usage-Based Pricing: The End of Subsidized AI Coding
GitHub Copilot is moving all plans to usage-based pricing on June 1, 2026, as confirmed by Microsoft yesterday. Instead of fixed request limits, users will be charged based on the actual compute cost of the models they use. According to the announcement, an agentic Copilot session can now cost far more than a quick chat—and Microsoft has been absorbing those costs. Per the Wall Street Journal (Oct 2023), Microsoft was losing over $20/month per user on average, with some heavy users costing up to $80/month.
Key Details
- Pricing shift: Users on a $19/month plan will get $19 worth of tokens—no more unlimited requests. The old premium request model is deprecated.
- Agentic usage: The product evolved into an "agentic platform" capable of multi-hour autonomous coding sessions. These sessions consume significantly more compute than simple inline suggestions.
- Subsidy ends: Microsoft called this "an important step toward a sustainable, reliable Copilot business." Translation: they can no longer subsidize nearly 2 million users.
- User reaction: Many users on HN and social media are calling the product "dead" and "completely ruined."
Context: The Broader AI Pricing Crisis
Ed Zitron's analysis traces this to the fundamental economics of generative AI: nearly every service—Copilot, OpenAI, Anthropic, Perplexity, Lovable—sells subscriptions below actual inference cost. For instance, Lovable offers 100 monthly credits + $25 cloud hosting for $25/month; Anthropic's Claude Pro is $20/month. These prices rely on venture capital subsidies. As Zitron puts it: "AI's economics don't make sense." Salesforce's Agentforce charges $2 per conversation, which even enterprise budgets may not sustain.
What Developers Need to Know
- If you use Copilot heavily, expect your bill to match usage—no more flat-rate access.
- Consider switching to self-hosted models (e.g., via Ollama or vLLM) for cost control.
- Monitor your token consumption: typical agentic sessions can burn through monthly credits quickly.
The shift reflects a larger industry trend: the era of subsidized AI is ending, and costs will be passed to users.
📖 Read the full source: HN AI Agents
👀 See Also
Open-sourcing AstaBrief, the fast report-generation model in Asta
Ai2 released AstaBrief 8B, an open-weights model that turns a research question plus retrieved literature excerpts into a cited report. Fast mode averages 51.1 seconds per report vs 178.5 for Thinking mode, and the training data ships alongside the weights.
AI Subscriptions Need a Reliable Meter: A Call for Service Transparency
A Reddit post argues that AI subscriptions should provide a basic service receipt showing what model was actually served, reasoning effort, context handling, and any load management, drawing parallels to weights-and-measures norms.
The AI Operator: A New Role for Agentic Workflows
Rish Gupta argues AI operators will be the key role in orgs within a year, combining technical skills (Python, LLM APIs, agent frameworks) with business process understanding to automate repetitive, high-impact tasks.
AI SREs Resolve Routine Incidents, But Engineers Lose Touch With Their Systems
Sylvain Kalache argues that AI incident responders reduce practice for engineers, worsening response to complex incidents, citing the Ironies of Automation and aviation training parallels.