OpenAI finally took GPT-5.6 out of gated preview and shipped it to everyone, expanding it globally. Unveiled back on June 26 and locked behind government-vetted access for weeks, the family is now public — and it comes in three flavors that map neatly onto three different jobs.
Meet Sol, Terra and Luna
Sol is the flagship, tuned for coding, and its headline trick is an “ultra” reasoning mode that delegates work to subagents — essentially spinning up helpers to tackle a hard problem in parallel rather than grinding through it in one context. Terra is the mid-cost workhorse at $2.50 in / $15 out. Luna is the fastest and cheapest, built for high-volume, latency-sensitive work. It’s the same tiering logic everyone’s converging on: one model to think hard, one to do the bulk work, one to do it fast and cheap.
Terra vs Claude Sonnet 5: the real fight
Look at Terra’s pricing again — $2.50/$15 — and you can see exactly who it’s aimed at. That’s a direct shot at Claude Sonnet 5, which has owned the “frontier-adjacent agentic model at a sane price” slot since its launch. The cheap-frontier tier is now a genuine knife fight: Sonnet 5, GPT-5.6 Terra, and Grok 4.3 all clustering around the same price band, all promising most of the capability at a fraction of the flagship cost.
The AI tool stack actually worth paying for
One email a week. The models, tools and moves that matter, stripped of hype and filtered so you don’t have to drink from the firehose. Free, and you can bail anytime.
What to actually use
Rough guide, and test it against your own workload before committing: reach for Sol when you’re doing serious agentic coding and the extra reasoning earns its keep; run Terra as your everyday default and benchmark it head-to-head against Sonnet 5 on your actual tasks, because at these prices the “right” answer is whichever one wanders off-plan less on your workflows; and route the boring, high-volume stuff to Luna. We already looked at how OpenAI plans to run Sol on Cerebras at 750 tokens/sec — if that lands broadly, Sol’s speed problem disappears too.
The takeaway
GPT-5.6 doesn’t blow up the leaderboard so much as it makes the mid-tier brutally competitive. For once, the interesting model isn’t the flagship — it’s Terra, and the pressure it puts on everyone else’s pricing. Good news if you’re the one paying the bill.