GPT-5.6 LAUNCHES AS SOL, TERRA & LUNA — LIMITED TO 20 ORGS AT US GOV REQUESTNEW MODEL
THE STORY
OpenAI began a limited preview of GPT-5.6 on June 26, 2026, introducing three models: Sol (flagship), Terra (balanced), and Luna (fastest/cheapest). At the US government’s request, access is restricted to about 20 pre-approved organizations via the developer API and Codex coding tool only. Sol sets a new benchmark on Terminal-Bench 2.1 and outperforms GPT-5.5 on biology workflows while using fewer tokens.
Sources: OpenAI · TechCrunch
· · · NEXT STORY · · ·
#2
OPENAI & BROADCOM UNVEIL JALAPEÑO — FIRST CUSTOM AI CHIP, BUILT IN JUST 9 MONTHSHARDWARE
THE STORY
OpenAI and Broadcom unveiled Jalapeño, OpenAI’s first custom AI inference chip, on June 24. The chip was developed from design to manufacturing tape-out in just nine months — with OpenAI’s own AI models helping to speed up chip design. Early results show roughly 50% better cost-per-inference than current Nvidia alternatives. Initial deployment is targeted for end of 2026, with gigawatt-scale datacenters built with Microsoft.
Sources: TechCrunch · CNBC
· · · NEXT STORY · · ·
#3
ANTHROPIC ACCUSES ALIBABA’S QWEN OF LARGEST-EVER DISTILLATION ATTACK ON CLAUDESECURITY
THE STORY
Anthropic sent a letter to the US Senate accusing operators affiliated with Alibaba’s Qwen AI lab of the largest known distillation attack on Claude. Between April 22 and June 5, 2026, roughly 25,000 fraudulent accounts generated 28.8 million exchanges with Claude, targeting its agentic reasoning and software engineering capabilities. The data was allegedly used to train Qwen models without permission. Senators Hagerty and Kim are moving to add sanctions to defense legislation.
Sources: CNBC · Nikkei Asia