Current frontier modelsClaude Fable 5.1 · GPT-6 Astra
More intelligence
per dollar.
Save up to 70% on AI costs. One predictable monthly price.
Hermes Agent
OpenAI Codex CLIGemini CLIAny OpenAI-compatible SDKOpenClawKilo CodeCursorGitHub CopilotAny OpenAI-compatible SDKClineWindsurf
Aider
ContinueAny OpenAI-compatible SDKOpenCode
Oh My Pi
AmpAny OpenAI-compatible SDKSee how much you can save.
How Standard Compute could optimize your workload
- Claude Fable 526%
- Kimi K321%
- GLM 5.220%
- MiniMax M318%
- DeepSeek V4 Flash15%
Fully customizable — you choose which models the router can use.
Your data, your region.
We select for providers that don't train on your prompts or retain them. Traffic stays inside the region you choose.
A clear budget. A fixed monthly bill.
One flat monthly price — and every individual plan starts with more compute than you pay for.
- $30 compute budget your first month
- Every model, smart-routed
- One agent at a time
- 1 API key
- $61 compute budget your first month
- Run agents in parallel
- Shared execution pool
- $142 compute budget your first month
- Priority scheduling
- Higher-capacity pool
- $403 compute budget your first month
- Highest priority scheduling
- $549 monthly compute budget
- Unlimited API keys
- $1,119 monthly compute budget
- More capacity for parallel agents
- $2,849 monthly compute budget
- Priority support
- Direct line to our engineers
- Compute sized to your volume
- Flat price for your whole team
- Uptime SLA & custom terms
- Consolidated invoicing
Create your company for free. Assign each member a plan from $19/month. Your team can mix plans.
Kudos guys! Really impressed with the model quality. Code output is better… Went from about $200 a month to just using your Standard plan. You really set a new standard for what compute should look like.
Side by side.
| Pay as you go | Subscriptions | Standard Compute | |
|---|---|---|---|
| A budget you can see | ~ | ✕ | ✓ |
| No surprise cut-offs | ✓ | ✕ | ✓ |
| A bill that never surprises | ✕ | ✓ | ✓ |
| Smart routing | ~ | ✕ | ✓ |
| Best provider rates | ~ | ✕ | ✓ |
| Unused budget given back | ✓ | ✕ | ✓ |
| More compute than you pay for | ✕ | ~ | ✓ |
Where the extra intelligence comes from.
- routeSmart routing
Every request is sized up and sent to the right model — efficient models for easy work, frontier intelligence when it matters.
- sourceProvider optimization
The same model runs on many providers. We route each request to whichever is fastest and cheapest right now.
- poolSubscription economics
Not every customer runs hot at once. Unused capacity in the pool becomes extra value for everyone in it.
Leave your agent running overnight. Wake up inside budget.
Turn on Smart pacing to spread your monthly compute budget across the month. When usage runs ahead of that pace, responses slow down, then recover as usage eases. Your monthly budget still applies: requests stop when it is used, until renewal or an upgrade. Watch the balance live in your dashboard. There are no automatic overage charges.
The internet did it. Streaming did it. Compute is next.
You get 2–7× more compute per $.
Three mechanisms stack in your favor.
Every frontier lab.
Always current.
The leading models, through one API. Each one editorially reviewed and routed across providers for price, speed, and quality.
Frontier
For demanding work- AnthropicClaude Fable 5.1
- OpenAIGPT-6 Astra
- GoogleGemini 3.1 Pro Preview
Open frontier
Leading open-weight models- Z.AIGLM 5.3
- QwenQwen 3.8 27B
- MoonshotKimi K3
Efficient
For everyday agent work- GoogleGemini 3.8 Flash
- QwenQwen 3.8 Flash
- DeepSeekDeepSeek V4 Flash
A selection from our model lineup. Availability evolves as models and providers are reviewed.
Two minutes to switch.
- Change the base URLapi.stdcmpt.com/v1
- Paste your API keyfrom your dashboard
- Set the model"standardcompute"
const client = new OpenAI({
baseURL: "https://api.openai.com/v1", baseURL: "https://api.stdcmpt.com/v1", apiKey: process.env.OPENAI_API_KEY, apiKey: process.env.STANDARD_COMPUTE_KEY,});
await client.chat.completions.create({
model: "gpt-5.2", model: "standardcompute", // the router takes it from here messages,
});Boring, the way infrastructure should be.
Common questions.
How does the monthly compute budget work?
Every plan states its budget up front, and you can watch it live in your dashboard. Requests run at full speed until the budget is used, then they stop until your period renews. No per-token billing, no overage charges, no surprise invoice. If you’d rather slow down than stop, optional pacing spreads what’s left across the month. The exact policy is on the fair-use page.
How can it be this much cheaper than what I pay now?
Three mechanisms, no magic. Most agent traffic is routine, so smart routing sends it to efficient models and saves 10x or more on those requests. The same model usually runs on several providers, and we buy from whichever offers the best rate at that moment. And unused budget in the shared pool goes back into serving members instead of expiring. It’s the same boring math your ISP runs, just pointed in your favor.
Do cheaper models mean worse results?
Smart routing chooses among the models in your selection for each request. You can include frontier models such as Claude Fable 5.1 and GPT-6 Astra for demanding work, alongside efficient models for routine steps. For more control, choose exactly which models the router may use, or select a single model. Compare results on your own tasks when changing your lineup.
What happens when my budget runs out?
Requests stop, and it never sneaks up on you. The dashboard shows exactly how much you have left at any moment. If you need more, an upgrade takes effect immediately; otherwise everything resets when your period renews. Nothing ever bills on top of your plan price.
Will it work with my agent?
If it speaks the OpenAI API, yes. Nearly every agent does. Three edits: point the base URL at api.stdcmpt.com/v1, paste your key, set the model to "standardcompute". Most people are done in about two minutes, and there are step-by-step guides for every major agent.
Is my data used to train models?
Standard Compute does not use your prompts or outputs to train models. The Regions & privacy page shows whether no-training and zero data retention enforcement are active for your keys. When active, both are required for upstream routing. Zero data retention covers prompt and response content; temporary provider caches may still apply, and service metadata is retained. Details on the Data & Privacy page.
Which models do I get?
Choose from frontier models including GPT-6 Astra, Claude Fable 5.1, Claude Opus 5, and Gemini 3.1 Pro Preview; open-weight options including GLM 5.3, Qwen 3.8 27B, Kimi K3, and MiniMax M3; and efficient models such as Gemini 3.8 Flash, Qwen 3.8 Flash, and DeepSeek V4 Flash. Select up to five models in your dashboard, or keep the default smart-routing lineup. New models become selectable after provider checks; adding a model does not change your saved selection.
What if it's not for me?
You’ll know before you pay. Signup gives you free usage with no credit card, so you can test every claim on this page first. Paid plans carry a 7-day fair refund: cancel in the first week and you only pay for the compute you actually used. After that, cancel anytime from the billing portal and your plan simply runs to the end of the period.