The billing switch, the community backlash, and what Project Polaris means for your costs.
JUNE 2ND, 2026 · BY JONATHAN HILDEBRANDT
GitHub flipped every Copilot plan from flat-rate to token-based AI Credits on June 1. Developers running heavy agentic sessions are projecting monthly costs above $750 — up from the old $29/month flat-rate subscriptions. The community discussion thread collected 900+ downvotes before the switch went live, and GitHub simultaneously removed the GPT-4.1 model fallback that teams used to cap spend.
At Build 2026 (June 2), Microsoft announced Project Polaris — an in-house coding model on custom Maia AI accelerators replacing GPT-4 Turbo as Copilot’s default in August. Polaris outperforms GPT-4 Turbo on HumanEval and MBPP benchmarks (per the Build 2026 recap) and runs cheaper on Microsoft’s own silicon. Until August, Cursor Pro at $20/month flat remains the clearest cost-certain alternative.
Anthropic submitted a confidential S-1 to the SEC on June 1, becoming the first major frontier AI lab heading to public markets. The $965B post-money figure follows a $65B Series H and surpasses OpenAI’s disclosed valuation. Share count and price are not yet set. For developers building on Claude, the filing signals the company is betting its near-term revenue can sustain a public-market debut.
Windows Agent Framework (WAF) is MIT-licensed. The Windows Agent Runtime gives agents first-class OS-citizen status. The new Windows Agent Store offers 85% developer revenue share — higher than Apple’s 70/30 split. If you’re building agents that target Windows end-users, this is a distribution channel worth watching.
Fleet mode lets Copilot CLI operate autonomously across multiple repos without per-step confirmation. Autopilot mode runs background tasks on bounded issues without a developer present. No separate pricing — both ship with existing Copilot tiers. Relevant for platform teams managing microservice fleets or monorepos.
Google DeepMind CEO Demis Hassabis told Axios he now sees 2029 as possible for AGI, citing growing confidence the industry has found the right technical path. He called the current agentic era “a little bit like a practice run” for more powerful systems ahead. His framing of 2029 as a possibility — not a certainty — is a notable public calibration from DeepMind’s CEO.
Opus 4.8 keeps the $5/$25 per million token pricing. The upgrade is free at the API level: one model-ID swap. Our review covers what’s new in reasoning and where Sonnet 4.5 still wins on cost-per-task. Read it.
Token billing is live today. We pulled the current rates and ran the per-persona math: when does $20/month Cursor flat-fee beat Copilot’s variable metering? The answer depends on how many agentic sessions you run per day. Read it.
Zapier has two integration surfaces now. MCP works for chat agents; SDK works for coding agents. The guide maps each to specific use cases so you pick the right one on the first try. Read it. Try Make.
n8n bills per execution. Zapier and Make bill per step. Which model costs less depends entirely on your workflow shape. We break down the math by complexity tier. Read it. Try Make.
Memory gained repo-level controls and CLI commands. Every context call is now billable under the new AI Credits system, so knowing how to scope Memory matters more than it did last week. Read it.
Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.
