Rosetta Intel

Product, model, and security developments across OpenAI / Anthropic / Google. 🛡 = security-related.

OpenAI

GPT-5.6 model family (Luna / Terra / Sol)— a new three-tier flagship release, priced at $1–$5 per million input tokens with a unified 1M-token context window; OpenAI says it surpasses Claude Fable 5 on long-horizon agentic benchmarks. AIToolsRecap· ThursdAI
IPO preparation— OpenAI is reported to be preparing a confidential IPO filing within weeks, with a listing as early as September and a private valuation of $730B; it has also proposed ceding a 5% stake to the U.S. government. Fortune
Lockdown Mode (retrospective)— on February 13 OpenAI shipped Lockdown Mode for ChatGPT and publicly conceded that prompt injection in AI browsers "may never be fully fixed"; it remains a focal point of agent security discussion this month. Securance

Fable 5 / Mythos 5 restored worldwide— supply resumed globally after U.S. export controls were relaxed on July 1, with a new cybersecurity classifieradded as a safety guardrail (availability had been suspended on June 12 over jailbreak concerns). Anthropic News· Releasebot
Sonnet 5 released— approaching Opus 4.8 performance, with introductory pricing of $2/$10 per million tokens (through August 31).
Claude for Teachers— free for certified U.S. K-12 teachers, with teaching skills and curriculum connectors included.
Talent and compute— Andrej Karpathy joined Anthropic to work on frontier LLM research; the company signed a $19B data center lease with TeraWulf.

Gemini 3.5 Pro (released July 17)— the 2.5 Pro architecture was abandoned for a wholesale rebuild, introducing a 2M-token context window, a Deep Think Reasoning Layer, and autonomous workflow capabilities, positioned against GPT-5.6 and Fable 5. TechTimes· The Agent Report
Talent attrition fallout— during the release delay four top researchers including Noam Shazeer and John Jumper left for OpenAI / Anthropic, and Alphabet's market value fell by roughly $225B at one point.
Gemini 3.5 Live Translate / Nano Banana 2 Lite / Omni Flash— real-time speech-to-speech translation (70+ languages, preserving tone of voice), the fastest and cheapest image model, and the Omni Flash API in public beta. Google Blog

Source configuration: see intel/sources.yaml