OpenRouter Free Models ↗
Explore free model endpoints through one familiar API interface.
Free tier
FREEAI GO · INDEPENDENT DIRECTORY
Free AI Models: Compare APIs, Downloads & Limits
Compare free AI models for text and images. Find API addresses, downloadable weights, account requirements, usage limits, and official model cards.
Need a free AI model? Use a hosted free API if you want to build without managing hardware. Download model weights if you want local control and can supply compute. For a browser conversation, start with the chat guide.
Free AI models: API access or a local download?
Official sources checked . A curated snapshot, with provider-reported limits. Availability and terms can change; these are not output-quality benchmarks or live availability tests.
7 options shown
| Model / provider | Useful for | Free access & conditions | Official sources |
|---|---|---|---|
OpenRouterHosted APInvidia/nemotron-3-ultra-550b-a55b:free | Text, coding, and agent prototypes; compare the provider’s current free variants. | Free variants: 20 requests/minute and 50/day when all-time credit purchases are below 10 credits. The documented higher tier is 1,000/day after purchasing at least 10 credits. The higher allowance involves a purchase. Account limits and upstream capacity apply; a negative balance can block free models. Check the chosen endpoint’s data policy. | Free model catalog ↗Official limits ↗API setup ↗ |
RequestyHosted APInemotron-3-ultra-550b-a55b | Text and coding workflows using an API gateway and its free model catalog. | 200 free requests per day, restricted to free models. The provider states no credit card is needed to start. Create an account and API key. The wider paid catalog is separate; select a model listed as free before calling it. | Free model catalog ↗Official limits ↗API setup ↗ |
Cloudflare Workers AIHosted API@cf/black-forest-labs/flux-1-schnell | Image generation with FLUX.1 [schnell], or text with a model from the Workers AI catalog. | A shared allocation of 10,000 Neurons/day, resetting at 00:00 UTC. Neurons measure compute; they are not a count of images, messages, or tokens. A Cloudflare account and API token are needed for REST access. Usage varies by model and settings. Some catalog models require a paid billing method; overage needs Workers Paid. | Official limits ↗API setup ↗ |
AIHubMixHosted APIIDs ending in -free | Subsidized text and image API variants in the provider’s free catalog. | Before a first top-up: 10 trial calls in total. A one-time top-up of at least $1 unlocks daily quotas: a shared 1M tokens/day and model-specific request caps. Daily quotas require that initial payment. Image models can have lower request caps. A $0 token price does not mean unlimited use or a permanent no-payment plan. | Free model catalog ↗ |
Qwen3-4BLocal downloadQwen/Qwen3-4B | A 4B-parameter text model for a first local chat or coding experiment. | Downloadable weights under Apache 2.0. Local inference has no hosted per-request fee; you supply the hardware and power. Choose a runtime-compatible format or quantization. Memory use depends on precision, context length, and runtime; “4B” is a parameter count, not a RAM requirement. | Official model card ↗Ollama model & command ↗ |
Gemma 3 4B ITLocal downloadgoogle/gemma-3-4b-it | Text and image inputs with text output, such as describing an image or summarizing text. | Downloadable open weights under Google’s Gemma terms. This is an image-understanding model, rather than an image generator. Hugging Face requires login and acceptance of the usage license for file access. Use a compatible multimodal runtime and check memory needs. | Official model card ↗ |
FLUX.1 [schnell]Local downloadblack-forest-labs/FLUX.1-schnell | Text-to-image generation through ComfyUI or Diffusers on your own hardware. | 12B-parameter downloadable weights, labeled Apache 2.0 on the model card. This applies to [schnell]; other FLUX variants have their own terms. Hugging Face requires login and acceptance of access conditions. Model files, runtime, and compute are separate from a hosted image API’s allowance. | Official model card ↗ |
API model IDs belong to their provider; a similar name on another provider may have different pricing and limits.
API addresses and what you need
The first three URLs are API base URLs. Cloudflare uses a model-specific REST endpoint; Ollama is a local server. Hosted calls need the provider’s account and API key or token. FreeAI Go supplies links, not keys or hosted inference.
- OpenRouter ↗
https://openrouter.ai/api/v1- Requesty ↗
https://router.requesty.ai/v1- AIHubMix ↗
https://aihubmix.com/v1- Cloudflare REST ↗
https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/{MODEL_ID}- Ollama · local ↗
http://localhost:11434/api
Keep hosted keys on your server. Match the model ID to the free catalog and check its input/output support before sending a request. A chat subscription does not automatically include API credits. Read the API guide →
Run a downloaded model locally
- Pick a model and runtime together. Qwen3-4B is a text option; Gemma 3 4B accepts images and returns text; FLUX.1 [schnell] generates images. Read the exact variant’s model card and access conditions.
- Check the download size, quantization, available memory, and intended context length. Start with a small task; model-file size alone does not describe total memory use.
- For text, install Ollama using its official instructions. Its Qwen3 4B library page lists the command below. The first run downloads the model and requires disk space and a network connection.
ollama run qwen3:4bDocumentation example; we have not benchmarked it on your hardware. Local API requests use http://localhost:11434/api/chat with a downloaded model and need no API key. Ollama cloud access has separate account and usage rules. Ollama setup ↗ · Local API documentation ↗
For images, follow the FLUX model card’s ComfyUI or Diffusers instructions. The Ollama text command above does not install an image-generation workflow. Compare image workflows →
Choose by the task you can verify
For an app prototype, compare API limits and behavior on the same short prompt. For offline work, choose downloadable weights and keep the runtime and integrations local. For images, compare generation settings and output rights. Run a small example before committing to a workflow; a usage ranking does not establish which model is best for your task.
Does free mean open source or commercial use?
Access price, downloadable weights, software licenses, and output rights answer different questions. Use the exact model card and the serving provider’s terms. The Apache 2.0 label on Qwen3-4B or FLUX.1 [schnell] does not describe every model in either family, nor does it make a hosted API free.