Standard Compute gives you access to a curated pool of leading models through a single API. You send one model name — standardcompute — and the router automatically matches each request to the best-fit model. Nothing to pick, nothing to configure.
Standard Compute routes across a curated pool of leading models from providers like OpenAI, Anthropic, and Z.ai (GLM). You never pick a model yourself — you send a single model name, "standardcompute", and the router automatically matches each request to the best-fit model.
The router weighs how hard the request is, current provider availability, cost, and speed, then serves the model that gives the best result for that specific call. Simple requests get fast, efficient models; hard multi-step problems get the strongest reasoning models. There is nothing to configure.
Requests are matched across a spectrum of model capability. You do not select these categories — the router chooses per request based on the work involved:
Deep reasoning — the hardest multi-step problems, deep analysis, and long-context work, served by the most capable models in the pool.
Coding & agentic — complex code generation and tool-heavy agent loops that benefit from stronger reasoning.
Balanced — everyday tasks that need a strong mix of quality and speed; the default for most automation workflows.
Fast & light — simple, high-volume, latency-sensitive work like classification, extraction, and short completions, served by fast, efficient models with very low latency.
Routing is transparent and fully automatic. You do not need to configure anything — send "standardcompute" and the system handles model selection for every request.
This also means you do not have to worry about individual provider outages. If one upstream provider has issues, the router automatically fails over to an alternative model of comparable quality, so a single outage does not break your workflow.
Responses are returned under the "standardcompute" model name; the specific underlying model and provider are abstracted away, so your integration never has to change.
Every plan includes unlimited requests and the same smart routing — you always get the full range of models. Plans differ by execution speed and scheduling priority, not by which models you can reach.
Starter ($9/mo) — simple access for experimenting with agents: shared execution pool with heavier batching and slower execution speed.
Standard ($39/mo) — balanced performance for everyday agent workflows at standard execution speed.
Fast ($99/mo) — faster execution and priority scheduling for active, complex agent workflows.
Turbo ($399/mo) — maximum execution speed and highest-priority scheduling for demanding, sustained agent automation.
You can change your plan at any time from the Dashboard.