regna-1 sends easy work to fast, cheap models and hard work to a blend of frontier models. Automatically. One API you never outgrow.
OpenAI-compatible. 1,000 free credits. No card.
Measured: 94.2% on RouterBench at $0.01 avg per question →Every request is graded first. Routine work runs on fast, low-cost models, and your bill follows the difficulty, not the hype.
Hard problems trigger Adaptive Blend: several frontier models run in parallel, get cross-checked, and are synthesized into a single answer.
Each request is metered by what it actually used and itemized in your console. No seats, no bundles, no mystery credit burn.
Every request is scored for difficulty the moment it arrives, before any expensive model is touched.
Routine work runs on fast, low-cost models and answers in seconds. Your bill follows the difficulty.
Hard problems trigger Adaptive Blend: several frontier models run in parallel, get cross-checked, and return one answer.
| Benchmark | regna-1 |
|---|---|
| MMLU (n=200) | 93.0% |
| ARC-Challenge (n=200) | 95.5% |
Measured end to end through the production API, 2026-07-13, routing overhead included. Read the full methodology →
regna-1 is OpenAI-compatible. Point your SDK at regnax.ai and you're done.
curl https://regnax.ai/v1/chat/completions \ -H "Authorization: Bearer $REGNA_API_KEY" \ -d '{ "model": "regna-1", "messages": [...] }'
from openai import OpenAI client = OpenAI(base_url="https://regnax.ai/v1", api_key=REGNA_API_KEY) reply = client.chat.completions.create(model="regna-1", messages=[...])
import OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://regnax.ai/v1", apiKey: REGNA_API_KEY }); const reply = await client.chat.completions.create({ model: "regna-1", messages: [...] });
ChatGPT Plus, Claude Pro, and Gemini stacked side by side run about $60 a month, flat, whether you used them or not. Regnax works the other way: frontier-class models behind one name, metered by what each request actually consumed. Light month, light bill. Start free with 1,000 credits and read your real number off the console.
1 credit = $0.01 of usage. Pay only for what each request uses.
regna-1 is a single model name backed by an orchestration layer. Regnax grades each request's difficulty, sends easy work to fast low-cost models, and blends several frontier models on hard problems, returning one answer.
Aggregators hand you a menu of hundreds of models and leave the benchmarking, choosing, and swapping to you. Regnax sells the finished answer instead: one model name, regna-1, that grades difficulty, routes or blends models, and returns a single response. And every request is itemized in your console, so you always know what a call cost.
Yes. Point any OpenAI SDK or HTTP client at https://regnax.ai/v1/chat/completions with model "regna-1" and a Regnax API key. Streaming, tool calling, vision, and up to 1M tokens of context are supported.
Usage is metered in credits: 1 credit equals $0.01 of usage, charged by what each request actually consumed. You start free with 1,000 bonus credits, and paid plans start at $14 per month.
No. Regnax evaluates new models and swaps them behind the stable regna-1 name, so your integration and prompts never change.