GPT 5.6 Arrives as Sol, Terra, and Luna
OpenAI split GPT 5.6 into three tiers, changed how caching is billed, and — the part that actually turns heads — said the flagship will run on Cerebras at 750 tokens per second.
TL;DR
- 🌞 Sol is the new flagship; Terra sits around GPT 5.5; Luna a touch below 5.4.
- ⚡ Sol on Cerebras next month at 750 tok/s — the standout detail.
- 💸 Prompt caching is now paid (~25% pricier to cache), but the cached read discount stays at 90%.
- 🚧 Limited to a handful of US partners for now; general availability promised in a few weeks.
The three tiers and pricing
GPT 5.6 ships as Sol (new flagship), Terra (roughly GPT 5.5 level), and Luna (slightly below 5.4). Per the announcement, on benchmarks it still falls short of the now-pulled Fable model.
| Tier | Roughly | Input / 1M | Output / 1M |
|---|---|---|---|
| Sol | New flagship | $5 | $30 |
| Terra | ≈ GPT 5.5 | $2.5 | $15 |
| Luna | Just below 5.4 | $1 | $6 |
The pricing model also changed: token caching is now paid — prompts that cache cost about 25% more up front — while the discount on reading cached tokens stays at 90%. Net effect depends on your cache hit rate.
The Cerebras angle
The most striking detail isn't the model — it's the silicon. GPT 5.6 Sol is set to launch on Cerebras next month at 750 tokens per second. At that speed, interactive agents stop feeling like batch jobs. Whether Terra and Luna get the same treatment — and at what price — hasn't been said.
Availability: GPT 5.6 is currently limited to a small set of US partners while OpenAI works through government approvals. General availability is promised in a few weeks — so treat the numbers above as the announced spec, not a verified, broadly-available product yet.
When it opens up
MegaBrain already routes OpenAI's models behind one OpenAI-compatible endpoint alongside 500+ others, with transparent per-token cost. When GPT 5.6 reaches general availability, you'll be able to switch to it with a one-line model-ID change and compare its real cost against Sol's open-weight rivals side by side. Watch the models page for the moment it lands.
Source
Announcement details via @ai_newz. Figures are as announced and not yet independently verified.
MegaBrain Gateway
500+ models. One API. No markup.
Use in Claude Code, Cline, Cursor, or any coding agent.
Newsletter
Stay in the loop
Get the latest model comparisons and guides — no spam, unsubscribe anytime.