Compute thresholds & AGI infrastructure stories

Jun 26, 2026

OpenAI Launches GPT-5.6 Models Under Restricted Government-Mandated Preview Due to Cybersecurity Risk Threshold

OpenAI has released its GPT-5.6 family (Sol, Terra, and Luna variants) in limited preview to select partners only, following a U.S. government request triggered by the models crossing cybersecurity risk thresholds in benchmarks. OpenAI publicly criticized the access restrictions as unsustainable and says they shouldn't become the norm.

Jun 26, 20267 sources

Apr 13, 2026

AI Industry Faces Compute Shortage with Outages, Rationing, and 50% GPU Price Surge

Surging demand for AI agents is colliding with limited compute capacity across the industry. Reports indicate Anthropic experiencing outages, OpenAI ending Sora access, and GPU prices jumping nearly 50 percent as infrastructure struggles to meet demand.

Apr 13, 20261 source

Apr 7, 2026

Anthropic Signs Multi-Gigawatt TPU Deal with Google and Broadcom Starting 2027

Anthropic has secured a deal with Google and Broadcom for multiple gigawatts of TPU computing capacity, set to come online starting in 2027, representing a major infrastructure commitment for AI model training.

Apr 7, 20261 source

Top claims

▪All three models in the GPT-5.6 family are classified at the High capability level for both cybersecurity and biological/chemical risk under OpenAI's Preparedness Framework.
▪On the Terminal-Bench 2.1 command-line coding benchmark, GPT-5.6 Sol Ultra scored 91.9%, outperforming Anthropic's Claude Mythos 5 at 88.0% and Google's Gemini 3.1 Pro Preview at 70.7%.
▪GPT-5.6 Sol costs $5 per million input tokens and $30 per million output tokens, Terra costs $2.50 and $15, and Luna costs $1 and $6.

Subtopics

Compute, chips & AI infrastructure2 AI compute & environmental impact1 AI governance1 AI policy1 AI Regulation1 AI security1 Large language models (LLMs)1

Related timelines