Short answer. Builder ($60) and Scale ($120) are unlimited fair-use plans: no per-token meter, with 10–16 concurrent requests and 600–1,200 RPM. Hobby ($8) and Starter ($20) publish daily caps of 2M and 5M tokens. Paid plans include all 39 models; the Free plan uses Micro and Lite.
What unlimited means
There are four things “unlimited” can mean. Plugsky uses the first two and publishes the rest:
- No per-token meter. Your invoice is the plan price, not a token count.
- Fair-use allowance. Builder and Scale are sized for normal production use, not for reselling capacity or running an unbounded scraper.
- Concurrency and RPM ceilings. A burst can queue even when tokens are not capped.
- Model context limits. The plan's request ceiling does not override the model's own context window.
If a provider promises unlimited tokens with no published concurrency, RPM or fair-use rule, you cannot forecast capacity. That is the claim to avoid.
Plans and actual limits
| Plan | Monthly | Requests/day | Tokens/day | Concurrent | RPM | Max input/output |
|---|---|---|---|---|---|---|
| Free | $0 | 100 new / 500 grandfathered | 50K new / 250K grandfathered | 1 new / 2 grandfathered | 30 | 32K / 2K |
| Hobby | $8 | 1,000 | 2M | 4 | 120 | 64K / 8K |
| Starter | $20 | 2,000 | 5M | 6 | 240 | 128K / 16K |
| Builder | $60 | Unlimited fair use | Unlimited fair use | 10 | 600 | 256K / 32K |
| Scale | $120 | Unlimited | Unlimited | 16 | 1,200 | 512K / 32K |
Annual billing is 20% off. New monthly customers get 30% off the first month, capped at $20. Paid plans include a 7-day money-back guarantee. Free-tier caps apply to accounts created from 2 November 2026; older accounts keep the previous allowance.
What is included
| Feature | What you get |
|---|---|
| Models | 39 models on paid plans, including fast, coding, reasoning and long-context tiers |
| API | OpenAI-compatible chat completions and embeddings; change the base URL and model name |
| Context | Up to 262,144 tokens on the main text models, subject to the plan request ceiling |
| Capabilities | Streaming, JSON mode, tool calling and embeddings on the relevant models |
| Deployment | Cloud, VPC, on-prem and air-gapped options |
| Billing | Flat monthly or annual; no per-token overage on self-serve plans |
If you depend on OpenAI-only surfaces such as audio, image generation, Assistants or fine-tuning, keep those on OpenAI for now. Plugsky is the text, vision and embedding path.
Which workloads fit
| Workload | Plan that usually fits | Why |
|---|---|---|
| Prototype and model comparison | Free or Hobby | Low daily volume; free covers Micro/Lite |
| Support assistant with steady traffic | Starter | 5M tokens/day; 6 concurrent requests |
| Multi-user agent product | Builder | Fair use, 10 concurrent, 600 RPM |
| High-concurrency agent fan-out | Scale | 16 concurrent, 1,200 RPM |
| Batch classification or extraction | Hobby or Starter | Daily token caps are predictable for scheduled jobs |
| Regulated or air-gapped | Enterprise | Private deployment and contract terms |
Agents usually hit concurrency before tokens. If you run 20 parallel tool calls, a 10-concurrent plan will queue half of them. Size for peak parallelism, not average tokens.
When flat beats per-token
At 2026 GPT-OSS rates of $0.15 input and $0.60 output per million tokens, an 80/20 input-output mix costs about $0.24 per million tokens. That gives a simple break-even:
| Plan | Monthly | Break-even |
|---|---|---|
| Hobby | $8 | ~33M tokens/month |
| Starter | $20 | ~83M tokens/month |
| Builder | $60 | ~250M tokens/month |
Below those volumes, per-token is cheaper. Above them, the flat plan wins until your model requirements force a more expensive model. The unlimited AI API guide has the full math and the questions to ask any provider.
Guardrails
- 80% warning. The API returns an
X-Plugsky-Quota-Warningheader at 80% of a daily cap. - 100% response. A HTTP 429 with an upgrade link, not a silent overage.
- Daily reset. Daily caps reset at 00:00 UTC.
- Fair use. Builder and Scale are for normal production use; reselling or unbounded automation is outside the terms.
- Concurrency. Requests above the concurrency ceiling queue or fail; test your peak before launch.
For a cost-aware agent that stays inside these limits, see the planner/executor example.
Frequently asked questions
Is Plugsky really unlimited?
Unlimited means there is no per-token meter on Builder and Scale. It does not mean infinite capacity. Those plans have a fair-use allowance and published concurrency and requests-per-minute limits. Hobby and Starter have explicit daily request and token caps.
What are the limits on Builder and Scale?
Builder is $60/month with 10 concurrent requests, 600 RPM and a 256K input / 32K output request ceiling. Scale is $120/month with 16 concurrent requests, 1,200 RPM and a 512K input / 32K output request ceiling. Both are fair-use plans, and the model's own context window still applies.
What happens when I hit a cap?
At 80% of a daily cap the API returns an X-Plugsky-Quota-Warning header. At 100% it returns HTTP 429 with an upgrade link. Fair-use plans can also queue or throttle at peak, so the concurrency and RPM ceilings matter even when tokens are not capped.
Is there a per-token charge on a flat plan?
No. Self-serve flat plans do not meter tokens on the invoice. You pay the monthly or annual plan price. The trade-off is that capacity is governed by the published caps and fair-use terms.
Can I run an AI agent on an unlimited plan?
Yes, but agents hit requests-per-minute and concurrency limits before token limits. Builder allows 10 concurrent requests and Scale allows 16. For fan-out agents, test a burst at 2–3× expected peak and watch for 429s and queueing.
Which models are included?
Paid plans include the full catalogue of 39 models. The Free plan uses the smaller Micro and Lite models. Main text models support up to 262,144 tokens of context, with streaming, JSON mode, tool calling and embeddings on the relevant models.
Can I deploy privately?
Yes. Plugsky offers cloud, VPC, on-prem and air-gapped deployment options. Private deployment is an enterprise arrangement, not a self-serve toggle, so confirm the model and SLA with sales.
Pick the plan that fits the ceiling
Start free, then move to Hobby, Starter, Builder or Scale when your daily tokens or concurrency demand it. No per-token overage on self-serve plans.
See plans → Start freeReviewed 20 November 2026. Plan limits were checked against the live pricing page. Fair-use terms and model context limits still apply; confirm current terms before committing.