
- Monthly visits
- 2,809
- 2026-10
- Monthly growth +2,986.8%
meraGPT - Task-specific models billed by token from prepaid credit
meraGPT offers three task-focused AI models through an API: Decider 1.1 returns probability distributions for typed questions about text or images; Restyler 1 rewrites machine-written prose; and Fanout 1 suggests web searches an AI assistant may run for a question. Use prepaid USD credit, with per-token rates and playground access.
- LLM Developer Tools
- LLM APIs
- typed decision API
- AI prose rewriting
- AI search query model




Capabilities
Typed decision distributions
Decider 1.1 answers typed questions about a state and returns probability distributions in one API call. Supply text or images and ask yes/no, choice among up to 255 labels, or rubric-score questions; it can return “I don't know” when offered and the state does not support an answer.
Machine prose rewriting
Restyler 1 rewrites model-generated prose to read like human writing. Send a document, including output from another model, to its endpoint; it is intended to preserve facts, meaning, names, numbers, links, and formatting, and has a 4,096-token context.
Likely search prediction
Fanout 1 returns web searches an AI assistant is likely to run before answering a question. Submit a question to its endpoint; it returns up to six search suggestions based on searches observed from ChatGPT, Claude, and Gemini.
Prepaid token API
The API charges prepaid USD balance according to actual input and output tokens at each model's listed rates. Add credit, create keys with optional model restrictions and spending caps, and call task-specific endpoints or use an OpenAI client pointed at meraGPT.
Pricing
meraGPT uses prepaid USD credit rather than a recurring subscription, and charges per actual token. Compare each model's input and output rates: Decider 1.1 has no output charge, while Restyler 1 and Fanout 1 do.
| Plan | Price | Included features and limits |
|---|---|---|
| Decider 1.1 | Input $0.04 / 1M tokens; output $0.00 / 1M |
|
| Restyler 1 | Input $0.04 / 1M tokens; output $0.16 / 1M |
|
| Fanout 1 | Input $0.04 / 1M tokens; output $0.16 / 1M |
|
The playgrounds have a daily allowance and can be used without an account; signing in spends prepaid credit instead. Optional auto top-up lets users set a trigger balance, top-up amount, and monthly ceiling.
Use cases
Classify and score incoming state
When software must interpret a conversation, alert, invoice, image, or agent trace, send the state and typed questions to Decider 1.1. It returns distributions for yes/no, label choices, or rubric scores in one call, which can support routing or threshold-based review.
Rewrite machine-generated documents
When a draft, documentation page, or assembled product copy needs more natural prose, send the document to Restyler 1. It rewrites the text while aiming to retain factual details, names, numbers, links, and formatting.
Explore likely assistant searches
When planning content for questions people may ask an AI assistant, submit a question to Fanout 1. It suggests up to six likely web searches, providing possible search directions rather than a guaranteed list of what an assistant will run.
Limitations
Restyler 1 has a 4,096-token context, so documents exceeding that model limit cannot be handled as a single context.
Decider 1.1 accepts choices of up to 255 labels; larger choice sets exceed its stated limit.
Fanout 1 returns likely search directions, not a fixed or guaranteed list; the FAQ says repeated runs on the same question can differ.
The Restyler benchmark described in the FAQ retains every fact in 41% of blocks and averages 73% of facts, so the provider says it is not a substitute for review where accuracy is critical.
Who it is for
Software developers integrating narrow AI tasks through an API
Teams maintaining generated-text pipelines that need prose restyling
Content planners exploring searches assistants may use before answering
FAQ
Which models are currently available?
Three models are listed as available: Decider 1.1 answers typed questions, Restyler 1 rewrites prose, and Fanout 1 suggests likely assistant searches. They share an API key and prepaid balance, with per-token billing.
What kinds of questions can Decider 1 answer?
It answers yes/no questions, choices among up to 255 labels, and rubric scores as probability distributions. Provide a state as text or an image and one or more typed questions; it can also answer “I don't know” when that option is offered and the state does not support an answer.
Does Restyler guarantee that every fact is preserved?
No. The FAQ reports that it keeps every fact in 41% of benchmark blocks and 73% of facts on average. The provider advises review when accuracy is critical; Restyler is not a summarizer or a general-purpose rewriter.
Does Fanout show exactly what an AI assistant will search?
No. It returns likely search directions, up to six suggestions for a question. The FAQ says repeated runs on the same question can have different query strings, so its output is not a guaranteed list.
How does meraGPT charge for API use?
Usage is billed per input and output token against prepaid USD credit, with model-specific rates and no per-request fee. Decider 1.1 has no output charge; $1 signup credit is available for seven days without a card, and the stated minimum to add credit afterward is $10 by card.
Comments