Microsoft's MAI Models Land in Copilot, Anthropic Files for IPO

<p><strong>Anthropic filed a confidential S-1 with the SEC</strong> at a $965 billion valuation and $47 billion revenue run rate, racing OpenAI to public markets. Meanwhile, <strong>Microsoft's MAI-Code-1-Flash</strong> rolled out to all GitHub Copilot plans — its first in-house coding model built without OpenAI data — as Copilot switches to token-based A...

Highlights

  • Anthropic filed a confidential S-1 with the SEC on June 1, racing OpenAI to public markets at a $965 billion valuation and a $47 billion annualized revenue run rate — up from $10 billion last year. (Anthropic)
  • Microsoft's MAI-Code-1-Flash is now live across all GitHub Copilot plans — its first in-house coding model trained without OpenAI data — while the reasoning sibling MAI-Thinking-1 enters private preview in Microsoft Foundry. (CNBC)
  • GitHub Copilot AI Credits token billing went live June 1, moving all plans to consumption-based pricing — a new budgeting variable for every Copilot-dependent team. (TechTimes)
  • Gemini 3.5 Pro, with a 2M-token context window and Deep Think reasoning, entered limited Vertex preview; Google committed to GA this month at I/O. (Google Cloud)

Key Signals

  1. Anthropic Confidentially Files IPO S-1 June 1, 2026

    Anthropic submitted a draft registration statement to the SEC following its $65 billion Series H, which pushed valuation to $965 billion. The company's revenue run rate has surged to $47 billion — over 4× last year's figure — led largely by Claude Code adoption in enterprise. The move positions Anthropic ahead of rival OpenAI in the public-market race and is the first potential mega-listing from a frontier AI lab. (Washington Post)

  2. Microsoft's MAI Models Cut the OpenAI Cord in Copilot Build 2026, June 2–3

    MAI-Thinking-1 — a 35B-active-parameter, ~1T-total sparse MoE with a 256K-token context window — was trained from scratch on commercially licensed data without OpenAI distillation. Microsoft claims it matches Claude Opus 4.6 on SWE-Bench Pro and scores 97% on AIME 2025; it enters private preview via Foundry. MAI-Code-1-Flash (5B params, optimized for speed) is rolling out now to all Copilot plans, with Project Polaris — the full-stack coding model — set to replace GPT-4 Turbo as the Copilot default in August 2026. (CNBC) (Neowin)

  3. Gemini 3.5 Pro Enters Limited Vertex Preview Announced Google I/O May 19, preview June 2026

    Targeting use cases previously served by Gemini Ultra, Gemini 3.5 Pro ships a 2M-token context window, Deep Think reasoning, and frontier multimodal capabilities. It shares the same SDK and configuration as the already-available Gemini 3.5 Flash, so teams can A/B the two tiers without code changes once Pro reaches GA. Enterprise access requires GCP contract allowlist enrollment. (Google Cloud)

Why It Matters / What To Watch

  1. Copilot operators need to recalibrate budgets now
    • GitHub Copilot AI Credits billing is live — all plans consume tokens, not seats. Review actual usage patterns before the first billing cycle closes to avoid sticker shock. (TechTimes)
    • MAI-Code-1-Flash is the new default model driving in-editor completions; test it against your team's actual workloads before assuming parity with the previous model. (Neowin)
  2. The frontier model tier is getting crowded — independent evals matter
    • Microsoft's SWE-Bench Pro claims for MAI-Thinking-1 are backed by Surge human raters, not external reproducible benchmarks; run your own evals before routing production agentic workloads to Foundry. (CNBC)
    • Gemini 3.5 Pro GA will complete Google's Vertex-to-Gemini Enterprise migration; teams building on the Gemini Enterprise Agent Platform should request Pro allowlist access now to begin side-by-side testing against Flash on cost-per-quality. (Google Cloud)

Quick Links