OpenAI has unveiled its next-generation GPT-5.6 model family, led by the flagship 'Sol' alongside 'Terra' and 'Luna.' The release is currently restricted to a small partner set following a request from the Trump administration to vet the systems for biological and cyber risks. Sol introduces 'Ultra Mode,' a capability that allows the model to coordinate autonomous subagents to handle multi-step workflows, and a 'Max Reasoning' toggle for deep analytical processing.
As raw intelligence between top-tier models like GPT-5.6, Claude 5, and Gemini 3.5 converges, the industry is shifting focus from model training to 'harness engineering.' This practice involves building sophisticated agentic loops, memory layers, and verification systems around the LLM, which can reportedly boost task success rates by up to 600% compared to zero-shot prompting.
SpaceX's $60 billion acquisition of the AI coding platform Cursor (Anysphere) this month is the largest startup acquisition in history, nearly doubling the previous record.
Which new GPT-5.6 feature allows the flagship model to leverage subagents to accelerate complex, multi-step work?