What is ScaledThought?
ScaledThought provides an OpenAI-compatible API for accessing open-weight language models, with account and key setup designed to be handled by an AI agent. It is aimed at developers and teams who want to use models such as MiMo-V2.6-Flash and DeepSeek-V4.1-Flash through an API or OpenAI SDK. Usage is charged per token from prepaid credits, with no subscription, and the service is hosted in the US.
ScaledThought Features:
OpenAI-Compatible API: Use the OpenAI SDK or plain curl with ScaledThought’s API, and switch between supported models by changing the model or base URL.
Agent-Led Setup: Claude Code, Codex, Cursor, or another agent that can run commands can create an account and API key from a setup prompt.
Live Open Models: The available models include Xiaomi’s MiMo-V2.6-Flash and DeepSeek’s DeepSeek-V4.1-Flash, with model prices available through the CLI.
Prepaid Token Billing: Calls draw on prepaid credits, with top-ups starting at $10 and per-token prices set separately for each model.
Key Spending Limits: Daily budgets and rate limits can be set per key, and API requests stop when the credit balance runs out.
Prompt Data Handling: Prompts and completions are processed in memory, not stored or logged, and are not used for training.
US Region Hosting: Accounts, keys, and usage records stay in the selected region, which is currently the US; Canada is planned.
ScaledThought Benefits:
Lower Setup Effort: Agent-managed signup and key creation can reduce the manual steps between choosing an API and trying a model.
Model Flexibility: Keeping an OpenAI-compatible interface makes it easier to try different listed models without rewriting an application around a new SDK.
Spend Control: Prepaid balances, key-level limits, and stopped requests at zero credit help teams keep usage within approved spending.
No Subscription: Teams pay for token usage from purchased credits instead of committing to a recurring subscription.
Reduced Data Retention: Not retaining prompts or completions can suit teams that want API processing without those conversations being kept by the service.
Use Cases:
Agent API Setup: Have a command-running agent create an account and key before using a supported model in a coding or research workflow.
Existing SDK Apps: Point an application already using the OpenAI SDK at ScaledThought’s API to send requests to its available models.
Model Evaluation: Compare the listed open-weight models on a task by changing the model used by an API request.
Budgeted Experiments: Test token-based model usage with a prepaid balance and set limits on keys used by agents or development projects.
Command-Line Testing: Use the CLI to inspect live models and prices, configure access, or make a test completion after adding credits.