Ask "which AI model is best" in 2026 and the only honest answer is "at what, exactly?" The field is multi-led and reshuffles quarterly; this site's blog covers the landscape as third-party editorial (see the 2026 model comparison and the routing architecture piece). Inside the product, that churn is exactly why routing is designed the way it is: x-rush.ai routes each task internally, by its shape, and treats provider names as plumbing that never reaches the interface.
What Smart Routing Means Here
Routing on x-rush.ai is a product behavior, not a settings panel. You pick a feature (Image Studio, Music Gen, AI Drama, and so on), describe the job, and the platform decides internally how to execute it. No model list is shown, no model picker exists, and no marketing promise ties any feature to a specific vendor's name. That is deliberate: a "supported models" list pasted into a UI would be stale within a quarter and would teach you to optimize the wrong thing.
What the routing decision never changes is what you are billed or what you keep. The contract below is stable even as the routing layer evolves.
What Routing Looks Like From the Outside
- Describe the job in the feature you opened. Workflow Copilot helps complete the generation brief; its help draws on an account-wide assistance pool (every 10 successful assists cost 0.5 credits).
- Confirm the quoted price. Every generation shows its fixed credit total before anything executes. Batches quote the full sum up front.
- The task runs. Routing, execution, and retries happen inside the task. You watch progress, not engine names.
- The result lands in your project. Every success is kept, grouped by generation batch, and a redo adds a version instead of overwriting. Nothing is auto-deleted.
What It Costs
Generation is billed in flat credits per output type, identical across features and plans: text 0.5, an image 1, audio 2, a music track 5, a video clip 10. The credit price does not swing with the task's difficulty or with anything that happens inside the routing layer.
Plans change capacity, not prices: Free refills 5 credits a day, Lite is $15 a month for 300 credits, and Pro starts at $29 a month for 800, with larger Pro tiers and credit packs above that. What a plan actually buys is concurrency, how many tasks run at once: 1 on Free, 2 on Lite, 3 on Pro.
Queue and Reliability
Tasks run in the order the backend accepts them. There is no priority queue attached to any plan; what a paid plan buys is the higher concurrent-run cap, not a faster lane. A global task center lets you see everything in flight and pause, resume, or cancel from one place.
Failure handling is part of the task, not your problem to orchestrate: a failed generation retries inside the same task, and if a result still cannot be delivered after retries, the reserved credits go back on your balance. A credit is only consumed by a delivered result.
Design Principles
- Providers are not product identity. Models are chosen by task fit inside the platform. Specific vendor and model versions are internal routing decisions, not user-facing claims.
- Capabilities appear when they are real. The engine catalog is maintained internally and changes quietly; features document what they do today, not what a roadmap intends. Where a capability is live, the feature page says so plainly; where it is not, the page says that too.
- Editorial stays separate. The blog reviews models from across the industry as editorial. A good review score on the blog is not a product announcement.
FAQ
Can I pick the model myself? No. Model choice is an internal routing decision and is deliberately not exposed. You control the feature, the job, the plan's concurrency, and the confirmation of every price; the engine behind the task is not part of the interface.
What happens when a better model ships? The routing layer can adopt it internally without any change to how you work: same features, same flat credit prices, same results flow. Because prices are per output type, upstream model changes never move your bill.
How is cost controlled if routing changes? By the flat per-output pricing above. Whether a task is easy or hard, a text unit is 0.5 credits and a video unit is 10, quoted before you confirm and refunded if undelivered.
Do paid plans get faster routing? They get more concurrent runs (2 on Lite, 3 on Pro versus 1 on Free), not a different queue. Tasks run in acceptance order for everyone.
Where do the model comparisons live? On the blog, as dated third-party editorial with sources and scheduled re-reviews. They explain what different kinds of tasks are worth; they are not a list of what the platform connects.