OpenAI Previews GPT-5.6: Sol Flagship With Subagent Ultra Mode, Terra and Luna Join the Family

TL;DR: OpenAI launched a government-coordinated limited preview of GPT-5.6 today — Sol (flagship, $5/$30 per MTok) ships "max" reasoning and an ultra mode using subagents; Terra ($2.50/$15) matches GPT-5.5 at 2x lower cost; Luna ($1/$6) is the new speed/budget tier.

OpenAI announced three new models on June 26, rolling out to ~20 trusted partners with government approval before a broader launch in the coming weeks:

Sol is the new flagship. It adds a max reasoning effort for deep analytical problems and an ultra mode that delegates complex work to subagents — similar in concept to Claude Code's multi-agent orchestration. Strongest coding, biology, and cybersecurity performance in the GPT line. Developer-customizable confirmation policy lets you define which high-risk API actions require user sign-off before execution.

Terra targets the everyday workload. OpenAI says it matches GPT-5.5 performance at half the cost ($2.50/$15 vs. $5/$30). Practical default for teams that don't need Sol's ceiling.

Luna is the fast, low-cost tier at $1/$6 — OpenAI's most affordable option with "strong capability" for high-volume throughput.

All three include improved prompt caching: explicit cache breakpoints, 30-minute minimum retention, 1.25x write rate, and a 90% read discount.

Why it matters: Pricing compression across the GPT line mirrors what's happened with Claude's model tiers — each new generation squeezes more capability into lower-cost models. Sol's ultra subagent mode is worth watching: if it reaches GA with a reliable API surface, multi-agent orchestration becomes a first-class feature across all major providers, not just Claude Code workflows.

Sources: OpenAI GPT-5.6 Sol preview | 9to5Mac coverage | GPT-5.6 System Card