Frontier Preview Mode — Price Frontier Models Before They Ship | Cromus
Frontier Preview Mode lets teams weigh the cost and effectiveness of frontier AI models on their workflows before those models reach general availability. Frontier models ship faster than any budget can react, and each one quietly redraws the cost curve. Instead of waiting for general availability, wiring a new model in, running real traffic, and reading the invoice a month later, Frontier Preview Mode answers the question up front: if I move this workflow to the new model, what happens to my bill, and is the quality jump worth it?
The frontier model Cromus tracks today is Anthropic's Claude Mythos 5 ($10 in / $50 out per 1M tokens), which moved from the Project Glasswing restricted set to an application-based open preview on July 15, 2026 — still not generally callable, and tracked for modeling only. OpenAI's GPT-5.6 family — Sol ($5 / $30), Terra ($2.50 / $15), and Luna ($1 / $6) — graduated to general availability on July 13, 2026, and Anthropic's Claude Fable 5 on July 1, 2026 (export controls lifted); all now price live from the model registry, so they have left this preview lineup. Pricing is provisional and mirrors each provider's published preview rates.
Frontier Preview Mode is a modeling layer, not a deploy path. A preview model only enters a simulation when explicitly pinned to a node; it is never auto-selected as the cheapest option, so a not-yet-shippable model can never quietly become the default. Every node priced on a preview model is flagged as provisional, and any workflow containing one carries a blocked-deploy banner until a generally available model is swapped back in. Frontier Preview Mode is rolling out on the Founding plan and above.