Key facts
| Router model | model="plugsky-fusion" blends tiers with no orchestration to build (live) |
| Default chain | Runs micro to pro to max, escalating on hard prompts |
| Strategies | cost_saver, balanced (default), max_quality and custom rules |
| Free plan | Two free AI models — plugsky-micro and plugsky-lite — with no card |
| Trial | 14-day full-access trial for paid-tier models and features |
| Cost shape | Flat self-serve plans; no per-token charges, fair-use RPM per tier |
| Model access | 30+ models behind one OpenAI-compatible endpoint |
| Roadmap | Classifier routing /v1/plugsky/route (model=auto) is coming soon |
TL;DR
- Skip building an ensemble — one model string gives you fan-out and escalation.
- Start on the free plan, benchmark paid tiers on the 14-day full-access trial.
- Keep flat pricing so growth does not create bill shock.
- Move to explicit strategies only when your data justifies them.
- Stay OpenAI-compatible so you never have to rewrite to leave.
How it works, step by step
- Ship the default: switch your model string to plugsky-fusion and run your existing prompts.
- Validate quality on the free models first, then trial paid tiers against a fixed prompt set.
- Adopt cost_saver for development and batch jobs and balanced for production.
- Add one custom rule at a time — long inputs or tool-heavy requests are the usual first candidates.
- Watch the routing logs for two weeks before adding more rules; resist tuning without data.
- When scale demands it, move high-value paths to max_quality or a fixed frontier model.
Original data
Try it yourself
Open the LLM cost calculator →
Orchestration you do not have to build
The standard startup mistake is building the platform instead of the product. An in-house ensemble means a router, per-provider clients, fallback logic, cost accounting and evaluation harness — months of work before the first customer. Fusion collapses that into one model string: model: "plugsky-fusion".
The router fans out across tiers, picks the best result and escalates when prompts are hard. You get multi-model behaviour on day one and keep your team focused on the workflow your users actually pay for.
Burn that does not spike
Per-token billing is a trap for usage-heavy agents: every retry, tool call and summarisation pass adds cost, and a launch spike can arrive before revenue does. Plugsky self-serve plans are flat monthly with unlimited fair-use usage and no per-token charges, so your invoice shape stays stable while you find product-market fit.
The free plan includes plugsky-micro and plugsky-lite with no card, and the 14-day full-access trial opens paid-tier models when you are ready to benchmark. That sequencing keeps spend tied to evidence rather than ambition.
Growing into routing
Default balanced is a good place to start and often a good place to stay. As your traffic becomes predictable, add rules with clear markers: development environments to cost_saver, long documents to a long-context tier, a premium path for the customers who pay for it.
Because the endpoint stays OpenAI-compatible, none of that tuning locks you in. If you later self-host, move providers or add a fixed model for compliance, the integration survives. 30+ models sit behind the same endpoint, so capability grows without a rewrite.
Honest comparison
| Concern | Plugsky Model Fusion | Building an ensemble | Single cheap model |
|---|---|---|---|
| Time to value | One model string | Weeks to months | Immediate |
| Quality on hard prompts | Automatic escalation | Depends on your logic | Capped by model |
| Cost shape | Flat monthly, fair-use RPM | Per-token across vendors | Per token or flat |
| Model access | 30+ models, one endpoint | Each integrated manually | One model |
| Maintenance | Configuration | You own the router | None |
Frequently asked questions
What is the cheapest way to try it?
Use the free plan, which includes plugsky-micro and plugsky-lite with no card, then start the 14-day full-access trial when you want to benchmark paid tiers.
Do I need to change my code?
Only the model string and base URL. The API is OpenAI-compatible, so your SDK and framework code stay as they are.
Does fusion cost more than a fixed model?
Self-serve plans are flat with unlimited fair-use usage and no per-token charges, so routing decisions do not change your monthly fee; see the live pricing page for current plans.
How much latency does escalation add?
Fan-out and escalation can add overhead on hard prompts. Use cost_saver or balanced for interactive traffic and pin a fixed model where latency is critical.
Can I control which models are used?
Yes. Strategies set the bias and explicit model names bypass routing entirely. Custom rules fire in order, first match wins.
What happens when the trial ends?
You pick a plan or continue on the free tier. There is no automatic charge.
Will I be locked in if I grow?
No. The endpoint stays OpenAI-compatible and 30+ models share one API, so you can change tiers, self-host or migrate without rewriting your integration.