OpenAI launches GPT-5.6 Sol, Terra, Luna; Sol matches Fable 5 capability at roughly half the cost

OpenAI launched the GPT-5.6 family on July 9 in three tiers: Sol (flagship), Terra (balanced, everyday work), and Luna (cost-optimized). On Agents' Last Exam (agentic workflows across 55 professional fields), GPT-5.6 Sol scored 53.6, eclipsing Claude Fable 5 by 13.1 points. Critically: Terra and Luna outperform Fable 5 at around one-sixteenth the estimated cost, while Sol matches Fable 5's intelligence at roughly half the estimated per-token cost.

On the Artificial Analysis Intelligence Index (a broad measure spanning agentic work, coding, and science), GPT-5.6 Sol with max reasoning comes within one point of Fable 5 while completing tasks in 61% less time at approximately half the estimated cost. In coding specifically (Terminal-Bench 2.1), Sol sets state-of-the-art at 88.8%, using less than half the output tokens and roughly one-third the cost compared to Fable 5. Luna outperforms Opus 4.8 at one-quarter the estimated cost. Efficiency gained from harness engineering, not model retraining.

Sol introduces two new reasoning tiers: max (more time for deep reasoning) and ultra (coordinates four agents in parallel across complex tasks). ChatGPT Work, powered by GPT-5.6, launches with agentic task completion capabilities, now available to Pro, Enterprise, and Edu subscribers. Sol is currently in limited preview to trusted partners per government coordination; general availability promised in coming weeks. Cerebras partnership ships Sol at up to 750 tokens per second for select customers.

For architects: GPT-5.6 tiers the frontier cost-capability curve, making smaller models viable for high-throughput work at predictable margins. The efficiency thesis—matching top performance at half the cost—directly addresses the token cost crisis cited by Palo Alto and Palantir. Expect rapid adoption in production agent systems where margin matters, particularly in cost-sensitive enterprise segments.