NVIDIA made Nemotron 3 Ultra available on June 4 under a permissive license, releasing a 550-billion-parameter sparse mixture-of-experts model built on a Mamba-Transformer hybrid architecture (ChatForest). With approximately 55 billion parameters active per token, the model sustains over 300 tokens per second in throughput, supports a 1-million-token context window, and carries an estimated inference cost roughly 30 percent below comparable closed-source frontier models (AI Tools Recap). On the Artificial Analysis Intelligence Index, Nemotron 3 Ultra scored 48, placing first among open-weight models developed in the United States and second globally, with Chinese open models still leading the overall leaderboard (The Decoder). Weights and model files are available on Hugging Face, via NVIDIA NIM microservices, and on OpenRouter (Hugging Face), with NVIDIA positioning the release as the first open frontier model designed primarily for agentic deployment patterns (ChatForest).
Anthropic on June 1 confidentially submitted a draft Form S-1 registration statement to the Securities and Exchange Commission, initiating the review process for a potential initial public offering (TechCrunch, Anthropic); reporting has placed the target valuation near $965 billion, with a possible listing as early as October 2026 and Wilson Sonsini - the firm that handled Google’s 2004 IPO - advising on public-market preparation (NPR). Separately, Anthropic announced that starting June 15, Agent SDK workloads - specifically the claude -p non-interactive flag, Claude Code GitHub Actions, and third-party apps authenticating via the Agent SDK - will draw from a new dedicated monthly credit pool rather than standard subscription usage limits (TechTimes). Credit allocations are $20 per month for Pro subscribers, $100 for Max 5x, and $200 for Max 20x, billed at standard API list rates; once the pool is exhausted, automated requests halt entirely unless the user has enabled overflow billing, and unused credits do not roll over (The New Stack, DevTool Picks). Interactive Claude usage through the chat interface and Claude Code in the terminal is not affected by the change (The New Stack).