The Vibe Father
Buyer's guide
OpenAI launched the GPT-5.6 family on July 9, 2026 in three tiers — Sol, Terra, and Luna — and the natural question is which one to pick. The honest starting point: these are brand new, and every capability claim so far is OpenAI's own. No independent scores have been verified yet. Our own numbers land on the live benchmarks as they come in (VCI = SWE-bench 40 / Terminal-Bench 30 / LiveCodeBench 30). So this guide picks tiers by their positioning and — critically — by price, because the pricing is real even when the scores aren't. We'll frame the three by the job each is built for and run the heavy-month math so you can choose with eyes open.
Sol — the planner tier
Sol is the top of the family, priced at $5 per million input and $30 per million output. OpenAI positions it as the deepest-reasoning tier, and the price signals the same thing — it's the most expensive of the three by a wide margin. The role it's built for is the planner seat: architecture, hard multi-file reasoning, the ambiguous work where you want the strongest model in the room and you're willing to pay for one clean pass. You don't run your whole workload on Sol; you reach for it when the decision is expensive to get wrong. Remember the caveat: OpenAI claims Sol leads, but the verified numbers aren't in yet.
Terra — the builder tier
Terra is the middle tier at $2.50/$15 — half Sol's price. This is the workhorse, the builder seat: the model you point at the bulk of your day-to-day implementation once a plan exists. It's meant to be strong enough for real repo work while costing little enough to run at volume. For most teams, Terra is the model that would carry the most tokens: the one turning a planned change into shipped code, batch after batch. Again, positioning and price are all we have so far — its independent benchmarks are pending.
Luna — the scout tier
Luna is the entry tier at $1/$6 — a fifth of Sol's input price. It's the scout seat: fast reads, summaries, triage, cheap first drafts, the high-volume low-stakes work where you want throughput and a low per-token cost rather than maximum depth. You'd route reads and drafts to Luna, keep the harder building on Terra, and escalate only the genuinely hard reasoning to Sol. First look in our Luna first look.
The heavy-month math
Here's where the tiers separate in dollars. We cost a realistic heavy month at 50M input tokens and 10M output tokens — the profile of someone running agents daily. Every cell below is a price fact; the SWE column is deliberately blank because no verified score exists yet.
| Tier | Role | In / Out per M | Heavy month | SWE-bench |
|---|---|---|---|---|
 GPT-5.6 Sol | Planner | $5 / $30 | $550 | not verified |
| GPT-5.6 Terra | Builder | $2.50 / $15 | $275 | not verified |
| GPT-5.6 Luna | Scout | $1 / $6 | $110 | not verified |
Reading the math
The spread is stark: Sol costs $550 a heavy month, Terra $275, and Luna $110. Running your entire workload on Sol costs five times what running it on Luna does — and most of a real workload is scout-grade and builder-grade work that doesn't need planner-tier reasoning. That's the argument against defaulting to the top tier. The efficient pattern is routing: Luna for the high-volume reads and drafts, Terra for the bulk of the building, Sol reserved for the genuinely hard planning where one clean pass is worth $30 output tokens. Do that and your blended cost lands far closer to Terra's number than Sol's, while you still keep the top tier available for the calls that matter.
Which tier to pick
- Pick Luna if your job is mostly reading, summarizing, triaging, and drafting — high volume, low stakes, and $110 a month is the value seat. See the Luna first look.
- Pick Terra if you're doing real implementation at volume and want a workhorse that costs half of Sol — for most teams this is the default builder.
- Pick Sol only for the hardest reasoning and architecture, and reach for it selectively rather than as your everyday driver.
- Pick all three and route — the cheapest total bill comes from sending each task to the tier that clears it, not from picking one tier for everything.
The honest close
Until independent numbers land, the right way to choose among Sol, Terra, and Luna is by role and by the price math — because the claims are OpenAI's, but the dollars are real and the gaps between tiers are large. Don't overpay by defaulting to Sol; most of your work is scout and builder work, and routing across all three is how you keep the bill sane while still having the planner tier on call. That per-task routing is exactly what a bring-your-own-key harness like The Vibe Father makes easy — one deck, any tier, switched per task. For the full pricing breakdown see our GPT-5.6 family pricing guide, the role pattern in best model for each agent role, and the verified scores as they arrive at /benchmarks.
