On July 9, 2026 — roughly two weeks after its limited preview launch — OpenAI opened GPT-5.6 to all API customers. The three-tier family is now generally available:
The three tiers
| Tier | Model ID | Input / Output | Context | Positioning |
|---|---|---|---|---|
| Sol | gpt-5.6-sol | $5 / $30 | 1.05M | Frontier reasoning — max & ultra modes |
| Terra | gpt-5.6-terra | $2.50 / $15 | 1.05M | Balanced quality at half Sol’s price |
| Luna | gpt-5.6-luna | $1 / $6 | 1.05M | Fast, cheap, high-volume |
Sol retains two extended modes: max (deeper reasoning, 88.76% Terminal-Bench 2.1) and ultra (sub-agent orchestration, 91.91% Terminal-Bench 2.1 — the published record).
What the GA changes
- No more waitlist. The limited preview restricted access to ~20 government-vetted partners; all API tier 4+ customers can now call all three tiers.
- Tiered routing is the real product. The three models share an architecture but differ in depth and cost. OpenAI explicitly positions this as “route by difficulty” — Sol for hard problems, Terra for routine, Luna for fast tasks.
- 1.05M context across all tiers. Every tier ships with the same 1.05M window — no context-gating by plan.
Why it matters
- OpenAI finally has a pricing ladder. GPT-5.5 was one model at $5/$30. GPT-5.6 is three price points covering the full spectrum from $1/$6 to $5/$30 — making OpenAI competitive on budget work for the first time.
- Terra is the quiet story. At $2.50/$15 it matches GPT-5.5 quality (per OpenAI’s benchmarks) at half the price — the volume pick for teams already on OpenAI’s stack.
- Sol ultra holds the terminal record. 91.91% Terminal-Bench 2.1 is the highest published score by any model. Combined with natively omnimodal input (text, image, audio, video), Sol is the clear CLI-agent leader.
The limited-preview backstory — government coordination under the June 2 executive order, the ~20-partner phase — is covered on the GPT-5.6 model page. For landscape context, see the 2026 LLM guide.