GPT-5.6 Sol nearly matches the top model at roughly a third of the cost and leads coding-agent evaluations in Codex. That changes the cost and quality tradeoff for agentic coding.
Key takeaways · AI-distilled
GPT-5.6 Sol (max) scores 59 on the Artificial Analysis Intelligence Index, 1 point below Claude Fable 5, at $1.04 per task, about a third of Fable 5's cost; Terra and Luna score 55 and 51 at $0.55 and $0.21 per task.
Across effort levels Luna and Sol always sit ahead of Terra on the intelligence versus cost frontier, so for any Terra setting there is a Luna or Sol setting that is smarter at the same cost or as smart for less.
Sol in Codex leads AA's Coding AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → Index at 80, topping DeepSWE and Terminal-Bench v2 and tying Grok 4.5 on SWE-Atlas-QnA, at about 40% lower per-task cost than Fable 5 in Claude Code.
GPT-5.6 is OpenAI's first family with cache-write pricing, at 1.25x the input price, while keeping the 90% cache-read discount. List prices are $5/$30 for Sol, $2.5/$15 for Terra and $1/$6 for Luna per 1M tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition →.
In AA-Briefcase Sol has the highest Presentation Elo of any model but ranks second to Fable 5, which leads on rubric score (56% vs 42%) and analytical quality Elo (1764 vs 1592).
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
