Forge Daily

MiniMax M3 Eclipses GPT-5.5 & Gemini 3.1 Pro — Opens the Frontier

First open-weight model to combine frontier coding (59% SWE-Bench Pro), 1M-token sparse attention context, and native multimodality — at $0.30/1M tokens vs Opus 4.7's ~$15.

Anthropic Files Confidentially for IPO at $965B — Fastest $1T in History

SEC draft registration filed June 1, days after closing $65B Series H. Q1 revenue $4.8B, Q2 projected $10.9B. Claude Code at $2.5B annualized. Founded 2021 — reached $1T in 5 years vs Google's 20.

Stanford AI Index 2026: Coding & Cybersecurity Agents Hit ~Human-Level

AI went from 60% to ~100% on SWE-bench in one year. Cybersecurity agent accuracy: 15% → 93%. Benchmarks saturating — harness quality now matters more than benchmark leadership.

AI Capital Barbell: 80% Flows to Top 10 — Mid-Stage Startups Squeezed

Cognition raised $1B+ at $25B pre (Devin), OpenRouter $113M Series B (CapitalG). Microsoft shopping for AI startups. Q1 2026 shattered records: $297B globally, 80%+ to AI.

Coding Agent Wars: Devin, Claude Code, Copilot, Cursor — $2.5B ARR Stakes

xAI Grok Build CLI launched (May 25), Grok Build 0.1 API (May 29) as Claude Code competitor. Mistral Devstral 2 (123B) on Bedrock at $0.40/1M input. Grok 5 still training on Memphis supercluster.

*estimated based on market comparable multiples