The top of the GPT-5.6 family, with new max and ultra reasoning modes — Ultra spawns internal sub-agents to break down hard problems before synthesizing an answer.
Sol keeps the flagship rate at $5 / 1M input and $30 / 1M output. For GPT-5.6 and later models, cache writes are billed at 1.25× the uncached input rate while cache reads keep the 90% cached-input discount.
OpenAI previewed the GPT-5.6 family — Sol, Terra, and Luna — on June 26, 2026, but shipped it first to roughly 20 approved partner organizations under a US government safety review. After the White House lifted those restrictions on July 8, GPT-5.6 launched publicly on July 9, 2026 through the API and Codex, with a ChatGPT rollout following. Sol is the flagship of the three, holding the same headline API rate as the GPT-5.5 flagship it replaces: $5 input and $30 output per 1M tokens.
The headline feature is a pair of new reasoning-effort levels — max and ultra — above the existing low/medium/high. Ultra decomposes a hard problem by spawning internal sub-agents, each handling a distinct part of the task before the results are merged. On Terminal-Bench 2.1, Sol Ultra reportedly scores around 91.9% and standard Sol around 88.8%, ahead of GPT-5.5 and competitive with the strongest frontier models.
Sol is built for the hardest agentic and reasoning work: long-horizon coding, tool-use loops, and multi-step planning. Ultra mode is the differentiator — it trades higher token cost and latency for materially better answers on problems that benefit from decomposition. For everyday tasks that don't need that depth, Terra delivers most of the quality at roughly half the price.
Prompt caching is more predictable in this generation: explicit cache breakpoints and a 30-minute minimum cache life make cost planning easier for long, stable system prompts.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| GPT-5.6 Sol | $5 | $30 | 1M |
| GPT-5.6 Terra | $2.50 | $15 | 1M |
| GPT-5.6 Luna | $1 | $6 | 1M |
| GPT-5.5 pro | $30 | $180 | 1M |
| Claude Fable 5 | $10 | $50 | 1M |
| Gemini 3.1 Pro | $2 | $12 | 1M |
Sol sits at the frontier flagship price point. For teams that don't need Ultra's decomposition, Terra is the value pick at half the cost; Luna is cheaper still for fast, high-volume work. GPT-5.5 pro remains OpenAI's extended-compute option at a much higher output rate.