OPENAI PREVIEWS GPT-5.6 (SOL/TERRA/LUNA) WITH NEW PRICING AND CACHE SEMANTICS UNDER LIMITED ROLLOUT
OpenAI previewed the GPT-5.6 model family (Sol, Terra, Luna) with new pricing and stricter prompt-caching rules in a limited U.S.-only rollout. Per OpenAI’s ow...
OpenAI previewed the GPT-5.6 model family (Sol, Terra, Luna) with new pricing and stricter prompt-caching rules in a limited U.S.-only rollout.
Per OpenAI’s own quote via Simon Willison, Sol, Terra, and Luna are in limited preview, with general availability planned in the coming weeks, and per‑1M‑token pricing set at $5/$30, $2.50/$15, and $1/$6 input/output respectively OpenAI quote via Simon Willison. The same source notes more predictable prompt caching: explicit cache breakpoints, a 30‑minute minimum cache life, cache writes billed at 1.25x input, and 90% discount on cache reads.
OpenAI’s community post confirms the new series and rollout posture OpenAI community post. A third‑party report adds claims about “maximum reasoning” and an “Ultra” subagent mode, pending official docs Interesting Engineering. If you’re planning deep, long‑context workflows, the GPT‑5.5 API explainer remains a useful guide to configuring effort, structure, and costs GPT‑5.5 API explainer.
Pricing and cache billing changed, which affects cost models, routing, and how you design long-context pipelines.
Tiered models enable cheaper defaults with on-demand upgrades, but rollout is gated and may delay production adoption.
-
terminal
Model routing: default to Luna or Terra and escalate to Sol for hard cases; compare quality, latency, and total cost per job.
-
terminal
Prompt caching economics: use explicit cache breakpoints and measure cache-write cost (1.25x input) vs 90% read discount on real workloads.
Legacy codebase integration strategies...
- 01.
Update cost alerts and budgets for new per-1M pricing and cache-write billing; re-run long-context job forecasts.
- 02.
Introduce a traffic splitter that routes by task complexity; keep GPT‑5.5 as fallback until GPT‑5.6 broad GA.
Fresh architecture paradigms...
- 01.
Design a tiered inference plan from day one: Luna/Terra for routine steps, Sol only where ROI is proven.
- 02.
Build prompts with deterministic schemas and cache breakpoints to reuse expensive context across steps.
Get daily OPENAI + SDLC updates.
- Practical tactics you can ship tomorrow
- Tooling, workflows, and architecture notes
- One short email each weekday