GPT-6 Sol and Luna: frontier AI at half the price changes the math for small teams

On the surface this looks like a routine model announcement — a new name and a leaderboard bump. The interesting part is the packaging. GPT-6 arrives as two variants tuned for opposite trade-offs: Sol leans into slow and careful reasoning for hard problems while Luna is optimized for latency and cost on everyday work. One model family with two gears instead of a single model pretending to fit every job.

Why the price cut matters more than the benchmark

Benchmarks move the conversation but pricing moves behavior. When frontier capability costs half of what it did last year the math changes for anyone building on top of it. A feature that was too expensive to serve at scale yesterday becomes a default capability tomorrow. That is not a small step — it reopens product roadmaps that were quietly closed for cost reasons.

The one-person studio wins first

The clearest winners are small teams. An independent developer or a two-person studio can now prototype against the strongest model instead of settling for the cheap one and hoping the idea survives the downgrade. In my own workflow the model bill had quietly become a design constraint — it now reads like a bill I can afford again. I document exactly how I build products this way on the FDE page.

How to use two models without overthinking it

• Draft and plan with Sol. Architecture decisions and tricky bugs deserve the slow gear.
• Ship with Luna. Chat replies and classification and summaries should feel instant.
• Re-evaluate monthly. The frontier moves and yesterday's split may already be stale.

The tooling you wrap around the model matters as much as the model itself — the same principle behind the product services I offer.