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

DeepSeek V4 pricing reality check: 178x cheaper cached tokens vs Opus, but capability lag acknowledged
DeepSeek V4 Pro input at $0.145/M tokens vs Claude Opus 4.7 at $5/M (34x cheaper); cache hits at $0.0036/M vs $0.625/M (173x cheaper). Capability is 3-6 months behind GPT-5.4 and Gemini 3.1 Pro.

Developer Switches from Cursor Composer 2 and Kimi 2.6 to Qwen3.6:35b-a3b for Enterprise Workloads
A developer reports using Qwen3.6:35b-a3b for daily work on a 500-700k LOC enterprise suite, citing better performance than Kimi 2.6 and DeepSeek 4 Pro/Flash, with costs ~$0.08/1M tokens on OpenRouter.

US Military Pressures Anthropic to Remove Claude Safeguards for Military Use
US military leaders including Defense Secretary Pete Hegseth met with Anthropic executives to demand removal of Claude's safeguards against military applications like mass surveillance and autonomous weapons. The Pentagon has given Anthropic until Friday to comply or face penalties including contract cancellation.

Pantheon-Reasoning-27B: A Dense Reasoning RP Model from Gryphe
Gryphe releases Pantheon-Reasoning-27B, an uncensored Qwen 3.6 27B fine-tune with full reasoning traces for roleplay. Built on Pantheon, Opus-4.6-Reasoning-24k, WorldSim, text adventure, and general RP data. GGUF quants available.