// MAGIC_LLM · AI_INFERENCE_API

One key. Every model. All the AI you need without picking a vendor.

Magic LLM is a curated reselling of the best open-weights AI models for every use case — chat, reasoning, code, embeddings, images, speech — behind a single OpenAI-compatible API. We pick the winners. You ship the product.

Pay per token from a $25-dollar credit pack. No monthly fee. No vendor lock. No multi-key sprawl.

No one-size-fits-all in 2026. Every shop is hybrid – multiple AI vendors, legacy systems, fast-moving tools, high-impact decisions. We help you navigate that. Need Anthropic or OpenAI specifically? Route them through Magic AI Gateway – same key, same ARC wallet, full analytics + caching + guardrails wrapped around your own keys.

Get a Magic LLM API key See the model shelf OpenAI-compatible · 60-second signup · cancel anytime

// TWO_PATHS

Curated shelf or your own keys. Same dashboard.

Magic LLM is a hub. Pick the path per workload - or use both. One signup, one dashboard, one ARC wallet for the management layer.

// MAGIC_INFERENCE

Curated open-weights shelf

Llama 3.3, Kimi K2.6, Gemma 4, Qwen Coder, Whisper, Flux. Per-token retail rates. We pick the winners + handle hosting. Best for: most workloads.

See the shelf

// MAGIC_AI_GATEWAY

Bring your own keys (BYOK)

Already have Anthropic / OpenAI / Gemini / Cohere keys? Route them through our Gateway for failover + observability + guardrails + team-key management. Your tokens stay on your provider's bill. We charge for the management layer only.

Set up the Gateway

// THE_SHELF

Top model per use case. Not all 50.

We've already done the work of figuring out which open-weights model wins for each task. Same key gives you all of them; the platform routes intelligently when you ask for a category, or you can pin a specific model.

Use case	Model	Best for	Retail
General chat	Llama 3.3 70B Meta (open-weights via Workers AI)	drop-in chat, agents, RAG completions	$1.50 per million tokens
Fast / cheap	Gemma 4 26B Google (open-weights via Workers AI)	high-volume light tasks, classification, summarization	$0.50 per million tokens
Reasoning	Kimi K2.6 Moonshot (frontier open-weights via Workers AI)	agentic planning, complex tool-use, frontier-grade reasoning	$2.50 in / $8.00 out per million tokens
Code	Qwen 2.5 Coder Alibaba (open-weights via Workers AI)	code completion, refactoring, code review agents	$1.25 per million tokens
Embeddings	BGE / GTE family BAAI / Alibaba (via Workers AI)	vector search, RAG indexing, semantic similarity	$0.05 per million tokens
Image generation	Flux / SDXL Black Forest Labs / Stability (via Workers AI)	hero images, product mocks, social posts	$0.10 per image generated
Speech-to-text	Whisper Large v3 OpenAI (open-weights via Workers AI)	transcription, meeting capture, call analytics	$0.01 per audio minute

// Catalog auto-syncs weekly. When upstream rates change, we update transparently. No "promotional pricing" theatre.

// PRICING

Pay per token. No monthly subscription.

Buy a credit pack. Use it on any of the 7 model use cases at the per-token rates shown above. Credits last 12 months. No monthly fee, no expiring quotas, no surprise multipliers. Bigger packs come with auto-refill so you never run out mid-request.

Starter pack

$25 credit

Lasts 12 months · auto-refill optional

First-key trial, light experimentation

Buy $25 pack

Builder pack

$100 credit

Lasts 12 months · auto-refill optional

Regular development + production traffic for solo builders

Buy $100 pack

Team pack

Scale pack

$1000 credit

Lasts 12 months · auto-refill optional

High-volume teams, agencies, growing apps

Buy $1000 pack

// CUSTOM_OR_NET_30

Need procurement-friendly invoicing or a custom credit pack? Email us - we'll set up NET-30 billing or a multi-year prepay at custom volume rates.

Talk to us

// Auto-refill: optional. Default fires when balance dips below $10 and tops up by $100. Low-balance email at $25. Same wallet (ARC - Automated Revenue Credits) covers Magic LLM, Magic Sites overage chat, Magic Loops runs, and Magic Agent calls - pay once, use anywhere.

// HOW_IT_WORKS

Drop in. Ship.

Sign up

Pick a tier, enter card. Your API key is live in 60 seconds.

Drop in

OpenAI-compatible. Set base URL to api.magicllm.auramediastudios.com/v1 + your key. Existing code keeps working.

Ship

Hit any of the 7 model use cases. We route, monitor, and bill against your bundle. Overage at retail.

// EXAMPLE

Three lines and you're live.

// Drop-in OpenAI client
import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.MAGIC_LLM_KEY,
  baseURL: "https://api.magicllm.auramediastudios.com/v1",
});

// Chat (Llama 3.3 70B by default)
const chat = await client.chat.completions.create({
  model: "magic-chat",
  messages: [{ role: "user", content: "What is sovereign AI?" }],
});

// Reasoning (Kimi K2.6)
const reasoning = await client.chat.completions.create({
  model: "magic-reasoning",
  messages: [{ role: "user", content: "Plan a 3-step research agent" }],
});

// Embeddings (BGE)
const embeddings = await client.embeddings.create({
  model: "magic-embed",
  input: "Magic LLM rocks",
});

// FAQ

Questions before you ship.

How is this different from buying directly from OpenAI / Anthropic?

We curate the open-weights models that actually win for each use case (Llama 3.3, Kimi K2.6, Gemma, Qwen, Whisper) and resell them with one bill, one key, one dashboard. No multi-vendor sprawl, no per-vendor key management, and we route to the model that fits the task automatically.

What's the margin / why is this a good deal?

Open-weights models on edge inference cost a fraction of frontier proprietary APIs. Llama 3.3 70B comes in around $1.50 per million tokens to you - that's 3-10x cheaper than GPT-4 / Claude Sonnet for most chat workloads. We add a small mgmt margin for the curation + routing + dashboard, but the headline rates already beat proprietary APIs.

Can I use my own Anthropic / OpenAI / Gemini key (BYOK)?

Yes - through Magic AI Gateway, our companion service inside the Magic LLM hub. Bring your Anthropic / OpenAI / Gemini / Cohere key, point your client at our Gateway endpoint, get failover routing + observability + guardrails + team-key management + usage analytics on top of YOUR keys. Magic Inference (this page) gives you our curated open-weights shelf at retail rates. Magic AI Gateway gives you a managed routing layer over your own provider keys. Same Magic LLM dashboard, same ARC billing for the management layer, your proprietary tokens stay on your provider's bill. See /magic-llm/gateway for setup.

How fast can I get a key?

Is this OpenAI-compatible?

Yes. Drop-in replacement for OpenAI client SDK. Set your base URL to api.magicllm.auramediastudios.com/v1 and your key, that's it. Existing code keeps working.

Does my data get used for training?

No. Inference-only. Nothing you send is stored beyond the request lifecycle, nothing is used for training, no third-party logging.

What about images and speech-to-text?

Same key, different endpoints. Flux / SDXL for image gen at $0.10/image. Whisper Large at $0.015/audio min. All billed against your monthly bundle.

Stop juggling vendor keys.

One bill. Every top open-weights model. OpenAI-compatible. Pay per token from a $25-dollar credit pack - no monthly fee.

Get an API key Talk to us first