OpenAI Releases GPT-5.6 (Sol, Terra, Luna): A Three-Class Model Family with a Tool for Calling the Response API
Compatible · OpenAI GPT‑5.6 · July 9, 2026
GPT-5.6 tier, cost and benchmark checker
All numbers below are taken from OpenAI's published GPT-5.6 tables and price lists. Move the controls to see how the category selection changes and the results.
Three categories of long lasting power
Generation number. Sol, Terra and Luna are stages that progress to their cadence. Select a category to see its role and rating card.
What an expensive workload indeed
Set the token volume for each request. Costs use OpenAI's published estimates per 1M. Cached entries save 90% off cached entries.
Average monthly spend · 30 days
Sol–1.0x
Terra–
Luna–
Payment warehouse: from GPT-5.6 onwards, cache writes are billed at 1.25x the rate of uncached entries. Cache reads maintain a 90% discount, with a minimum cache lifetime of 30‑ and clear cache breakpoints.
Benchmark scores, combined
Values are the published OpenAI table. A dash in that table means that the score was not reported, so the model is not included here. OpenAI says its latency and cost calculations are done offline, not measured in production.
GPT-5.6 family
GPT-5.5
Anthropic
Google
Ultra: four agents, by default
Ultra connects up to four agents automatically. It trades higher token usage for stronger scores and faster time to result. Change to compare with Sol's single agent base. OpenAI also charts 16 agent runs on BrowseComp and SEC-Bench Pro.
Points · one agent
Terminal-Bench 2.1
BrowseComp
SEC-Bench Pro
Where to find it: Ultra works on ChatGPT Work for Pro and Enterprise, and Codex for Plus and above. In the API, the multi-agent beta in the Answers API creates an ultra-like flow.
