Free Models, Zero Compromise: Routing to Local and Free Tiers

Not every request needs a frontier model, and a surprising share of them can run for nothing at all. The problem is that "free" usually sounds like "worse," so teams pay for every request just to be safe. Routing is what removes that trade-off. There are actually two separate pools of zero-cost inference, and they behave very differently. It's worth knowing both before you decide what to send where. Run a model on your own hardware and the marginal cost of a request is zero. Manifest connects local servers the same way it connects anything else: Ollama, LM Studio, and llama.cpp, plus any other OpenAI-compatible server you point it at. Three things make local special. It's free, in the sense that you pay for electricity, not per token. It's private, because the prompt never leaves your machine. And it has no rate limits, because you aren't sharing a quota with strangers. The catch is just as simple: you need the hardware, and a small local model is not Opus. Which is exactly why you don't send it the hard work. The second pool lives in the cloud. A lot of providers run a genuinely free tier, and Manifest keeps a curated list of them. At the last sync that was over a hundred free models across more than a dozen providers. A few highlights, all free and most without a credit card: Groq serves Llama 3.3 70B and Llama 3.1 8B on ultra-fast hardware, Cerebras pushes around 2,600 tokens a second, and OpenRouter exposes more than 35 models with a :free suffix, including DeepSeek R1 and Qwen3 Coder. NVIDIA NIM opens 100+ models to anyone in its developer program, while Google's Gemini 2.5 Flash and Mistral's free Experiment plan round things out. That catalog isn't static. It's an open-source list we maintain and sync once a day, so as free tiers appear and disappear the page keeps up. You can browse the whole thing at manifest.build/free-models. This is the honest part, because "zero compromise" is a claim worth earning. The compromise people are afraid of is quality, and that is the one routing actually removes. You don't send a free or local model your hardest request and hope. You send it the work it handles just as well as anything else: summarizing a ticket, extracting a field, classifying a message, drafting a first pass. For those tasks, a fast 8B model and a frontier model produce the same answer, and only one of them shows up on the invoice. Free tiers do come with strings, and we won't pretend they don't. They have rate limits, often a few dozen requests a minute and a few hundred a day. Some cap the context window on the free plan. And a few log or train on your free-tier traffic: Google notes that free Gemini prompts may be used to improve its products, and some trial keys aren't cleared for commercial work. Our list flags those warnings on each provider for exactly this reason. That is the whole case for handling this with routing instead of by hand. Non-sensitive, simple work goes to a fast free tier. Anything private stays on a local model, where nothing leaves your machine. And real frontier models are kept for the requests that genuinely need them. Each request lands where it fits, so the limits of any one option stop being your problem. You don't wire your app to Ollama, or to Groq, or to any single provider. You point it at one endpoint, set the model to auto, and let Manifest score each request and route it. Assign a local or free model to your simple and standard tiers, or list them as fallbacks, and the easy traffic stops costing money without you touching the code again. And it isn't a black box. Every response carries headers showing which model answered, which tier it landed in, which provider served it and why, so you can see exactly how much of your traffic ran for free. The point of "zero compromise" was never that free models have no limits. It's that those limits stop mattering once each request goes to the thing that handles it best. Sometimes that is a frontier model. Far more often than most teams expect, it's something that costs nothing. Manifest is open source, and the free-models catalog is live. Browse it and connect your first free provider at manifest.build/free-models.