Key facts
| Model class | Deepest-reasoning profile in the Plugsky catalogue |
| Best for | Multi-step analysis, difficult maths and high-stakes review |
| Context class | 128K-class window; live limits are published per model |
| Capabilities | Reasoning, streaming, function calling, JSON mode and long-context |
| Pricing tier | Paid-plan model; free plan covers plugsky-micro and plugsky-lite |
| API | OpenAI-compatible /v1/chat/completions — no SDK changes |
| Resilience | Backup upstream plus same-profile fallback peers |
| Product status | Live |
TL;DR
- The deepest-reasoning tier for problems that justify the cost.
- Use it to escalate and to benchmark other models.
- Keep it off interactive defaults — it is the slowest path.
- Ask for a concise conclusion; reasoning output is verbose.
- Validate quality on your hardest prompts, not on average chat.
How it works, step by step
- Read the live model card at /models for window, limits and feature flags.
- Collect your hardest real requests into an escalation evaluation set.
- Run plugsky-ultra and your current model on that set.
- Score outcomes objectively — tests, calculations or a rubric.
- Set escalation triggers and cap attempts per request.
- Cache premium outputs and review triggers when model cards change.
Try it yourself
Open the LLM cost calculator →
What plugsky-ultra is
plugsky-ultra is the deepest-reasoning profile in the Plugsky catalogue. It is built for questions that cannot be answered well by pattern matching alone: multi-step analysis, difficult arithmetic or logic, trade-off decisions and review of work where an error is expensive. It supports the standard OpenAI-compatible surface — streaming, function calling, JSON mode and long-context inputs — and is listed as 128K-class on the live model card.
Check /models for the current engine, limits and routing, since top-tier profiles are updated most often.
When to choose it
Choose plugsky-ultra as an escalation target and a reference point. It earns its place when a cheaper model fails validation, when the request is flagged as high-stakes, or when you need a quality baseline to calibrate your default models against. It is also a sensible choice for offline jobs where throughput is irrelevant but correctness is not.
It is the wrong choice for routine chat, extraction, tagging or anything interactive. Compare it with plugsky-max and plugsky-frontier on your hardest prompts — the top tier is a cluster, not a single winner.
Top-tier trade-offs
The deepest tier asks for patience and discipline in return for quality:
- Budget latency: it is the slowest path in the catalogue.
- Constrain output so reasoning does not crowd out the answer.
- Send complete context — an under-specified prompt wastes the call.
- Verify outcomes objectively rather than trusting fluent reasoning.
- Log escalations so you can see whether the extra quality changes results.
Self-serve plans are flat monthly with fair-use usage, so map escalation volume to plan limits on the live pricing page rather than estimating per-token spend.
How to switch to plugsky-ultra
Switching is one model name on the same endpoint: {"model": "plugsky-ultra", "messages": [{"role": "user", "content": "Compare these architectures, then recommend one with reasons."}]}.
Test it on your hardest real requests, compare the outcome with your current strongest model, and adopt it only for the paths where it measurably improves results.
Honest comparison
| Dimension | plugsky-ultra | plugsky-max and frontier | Workhorse tiers (e.g. pro) |
|---|---|---|---|
| Best fit | Deepest reasoning and high-stakes review | Hard analysis and synthesis | Everyday production work |
| Reasoning depth | Highest | Highest tier cluster | Strong general purpose |
| Latency profile | Slowest | Slowest tier | Balanced |
| Context class | 128K-class | 128K-class | 128K-class |
| Use as | Escalation and baseline | Escalation and top default | Default production model |
| Plan | Paid | Paid | Paid |
Frequently asked questions
Is plugsky-ultra free?
No — it is a paid-plan model. The free plan covers plugsky-micro and plugsky-lite with no card required, and a 14-day full-access trial lets you evaluate paid tiers first.
What is plugsky-ultra best at?
The hardest reasoning work: multi-step analysis, difficult maths and logic, architecture decisions and high-stakes review where a wrong answer is expensive.
Should I use it as my default model?
No. It is the slowest and most expensive path. Keep faster tiers for routine traffic and reserve ultra for escalations and quality baselining.
How is it different from plugsky-max?
Both are top-tier, but their design points differ. Benchmark them on your hardest prompts and choose per workload rather than assuming one dominates.
How do I know the extra quality is real?
Score outcomes objectively with tests, calculations or a rubric, and compare the escalation rate and cost per successful task against your default model.
Does it support tools and structured output?
Yes — streaming, function calling and JSON mode are part of the OpenAI-compatible surface, so existing tool-using code works unchanged.
What context window does it have?
It is 128K-class today, but the live window and output limit are published per model on the catalogue.
How is pricing structured?
Self-serve plans are flat monthly with fair-use usage rather than per-token billing. See the live pricing page for current plans.