OpenAI released GPT-5.6 Sol, Terra and Luna on July 9 across ChatGPT, Codex and its API. The company said the worldwide rollout could take as long as 24 hours. Rather than one replacement model, the launch presented three tiers intended to balance capability, speed and cost for different users and workloads.
OpenAI also introduced higher-compute Max and Ultra modes. It described Ultra as coordinating four agents by default. More compute and coordination can change completion time and cost, so a model name alone is not enough to compare two runs. The selected tier, mode and task setup all belong in any reproducible evaluation.
Initial API prices per million tokens were listed at $5 for input and $30 for output on Sol, $2.50 and $15 on Terra, and $1 and $6 on Luna. OpenAI revised pricing on July 30, with a large reduction for Luna and a smaller reduction for Terra. Current pricing should be taken from the live provider page, with the effective date shown next to any table.
OpenAI published performance and safety evaluations with the release. Those results are provider-run claims and should not be treated as an independent ranking against Claude, Gemini, Grok or earlier GPT systems. Launch-day coverage also examined the government's safety decision, adding policy context without independently reproducing the model benchmarks.
For users, the durable question is which tier completes a real job at an acceptable total cost and delay. A cheaper token rate can lose its advantage if a task needs more retries, while a premium mode may be unnecessary for routine work. Price, latency, tool use, context and supervision should be measured together.
What to watch next
Track current API prices and independent cost-per-task tests that disclose model tier, compute mode, retries, latency and tool settings.
Sources
The GPT-5.6 announcement and TechCrunch launch coverage were reviewed on Aug. 3, 2026. Provider benchmarks were not reproduced, and price tables require a fresh check before publication.
We link to primary documents and first-hand reporting whenever possible.