Free AI Models & Free Tokens - 47 Free LLMs and APIs (2026)
Explore free AI models from OpenRouter, OpenCode, Gemini, Groq, Cloudflare, Cerebras, Z.Ai, SiliconFlow, and ZenMux. Most work without a credit card. The list includes 47 free and trial models such as NVIDIA Nemotron 3 Ultra 550B, Gemini 4 Hydrogen, Gemini 3.8 Flash, DeepSeek V4.1 Flash Free, MiniMax M3.1 Flash Free, GLM-5.3-Flash Free, Ling 3.2 Flash, Particle-1, and Ming Image 0.1 Design. Most support standard OpenAI-compatible APIs; OpenCode’s free models are available only inside its client. Many platforms offer free tokens or a daily free quota.
📊 Quick Stats
- 47 free/trial models
- 9 providers
- $0 total cost
- 1M maximum context
🎟️ Free Tokens and Free Quotas
“Free tokens” means different things on different platforms: a sign-up credit, a daily or per-minute request quota, or simply no token charge for selected models. The table below summarizes the current public rules by platform:
| Platform | Free quota | Notes |
|---|---|---|
| OpenRouter | Small free credit for new users; :free models 50 requests/day (1,000/day after a $10 top-up) | Free models are not billed per token |
| OpenCode | Daily request limit not published; LongCat 2.5 100 RPD | OpenCode client only, not a general API |
| Gemini | Free tier rate-limited by model | Inputs and outputs are used to improve Google products |
| Groq | 30 RPM / 1,000 RPD | Free plan, extremely fast models |
| Cloudflare | 10,000 Neurons per day | Resets at 00:00 UTC |
| Cerebras | $5 trial credit (30 days) | Requires a verified payment method |
| Z.Ai | Free model quota | See the console for details |
| SiliconFlow | Free model quota | See the console for details |
| ZenMux | $0/$0 free models | See the platform rules |
Platform policies change. Refer to each platform’s official documentation and account page; this site does not promise a “monthly free token refresh”.
🔥 Popular Free Models
| Model | Provider | Context | Price |
|---|---|---|---|
| NVIDIA Nemotron 3 Ultra 550B | OpenRouter | 1M | $0 |
| Gemini 4 Hydrogen | Gemini | 1M | $0 |
| Gemini 3.8 Flash | Gemini | 1M | $0 |
| Gemini 3.6 Flash | Gemini | 1M | $0 |
| Gemini 3.5 Flash-Lite | Gemini | 1M | $0 |
| GPT-OSS 120B | Groq | 131K | $0 |
| Qwen 3.8 27B | Groq | 131K | $0 |
| GPT-OSS 120B | Cloudflare | 128K | $0 |
| Qwen 3.8 27B | Cloudflare | 262K | $0 |
| GLM-4.7-Flash | Cloudflare | 131K | $0 |
| NVIDIA Nemotron 3.5 Super VL | OpenRouter | 1M | $0 |
| Ling 3.2 Flash | OpenRouter | 1M | $0 |
| Particle-1 | OpenRouter | 1M | $0 |
| Recursive Flint 1.0 | OpenRouter | 262K | $0 |
| Liquid LFM3-1.7B | OpenRouter | 262K | $0 |
| Baidu ERNIE 5.5 Flash Preview | OpenRouter | 1M | $0 |
| Google Gemma 4 31B | OpenRouter | 262K | $0 |
| Google Lyria 3 Pro | OpenRouter | 1M | $0 🎵 |
| DeepSeek V4.1 Flash Free | OpenCode | 200K | $0 (client) |
| GLM-5.3-Flash Free | OpenCode | 200K | $0 (client) |
| MiniMax M3.1 Flash Free | OpenCode | 1M | $0 (client) |
| LongCat 2.5 Free | OpenCode | 1M | $0 (client) |
| Agnes 2.5 Flash (Free) | ZenMux | 512K | $0 |
| GLM 4.7 Flash (Free) | ZenMux | 200K | $0 |
| GLM 4.6V Flash (Free) | ZenMux | 200K | $0 |
| Ming Image 0.1 Design (Free) | ZenMux | 8K | $0 🎨 |
| Ming Image 0.1 Design Layer (Free) | ZenMux | 8K | $0 🎨 |
| GLM-4.7-Flash | Z.Ai | 200K | $0 |
| GPT-OSS 120B | Cerebras | 65K | $0 (trial) |
| Qwen 3.8 27B | Cerebras | 64K | $0 (trial) |
| Xing4.0 29B | SiliconFlow | 256K | ¥0 |
| Hunyuan-MT-7B | SiliconFlow | 33K | ¥0 |
| PaddleOCR-VL 1.5 | SiliconFlow | 16K | ¥0 |
💡 How to Use Free AI Models
- Choose a model - Pick a provider from the left navigation (OpenRouter / OpenCode / Gemini / Groq / Cloudflare / Cerebras / Z.Ai / SiliconFlow / ZenMux).
- Get access - Most platforms issue a free API key in the console. OpenCode’s free models need no separate API key; select them directly in the OpenCode client.
- Call the model - Most platforms use standard OpenAI-compatible endpoints. For OpenCode, use
opencode runor specify the model inopencode.json. - Start at zero cost - Every model starts at $0. Cerebras requires a verified payment method and provides a $5 trial credit (30 days).
🤖 For AI Agents and Programmatic Access
The site provides machine-readable endpoints for AI agents and scripts:
| Resource | URL | Purpose |
|---|---|---|
| JSON API | /en/models.json | Structured data for all 47 models: model ID, context length, input modalities, and price |
| llms.txt | /en/llms.txt | Site structure index for LLMs (llmstxt.org) |
| RSS | /en/index.xml | Feed of all model pages (sorted by update time) |
| Sitemap | /sitemap.xml | All page URLs and last-modified dates |
Example (fetch every model in one request):
curl https://freemodels.net/en/models.json | jq '.models[] | select(.context >= 500000)'Each record contains:
{
"id": "recursive/flint-v1.0:free",
"name": "Recursive Flint",
"provider": "openrouter",
"context": 262144,
"max_output": 235929,
"input_modalities": ["text"],
"price": { "input": 0, "output": 0, "currency": "USD" },
"free": true,
"url": "https://freemodels.net/en/providers/openrouter/recursive_flint-v1.0free_free/"
}💡 Each model page also embeds Schema.org JSON-LD with
modelID,contextWindow, andinputModalities, which search engines and AI parsers can extract directly.
❓ Frequently Asked Questions
Are these free AI models really completely free?
All listed models can be used starting at $0, and most platforms do not require a credit card. OpenRouter’s:free models are free but have daily request limits; Cerebras has no permanent free tier and requires a verified payment method for a $5 trial credit (valid for 30 days); Gemini’s free tier uses inputs and outputs to improve Google products; OpenCode’s free models are only available inside the OpenCode client. Limited-time free models can change at any time, so refer to each platform’s official page. The site currently lists 47 free/trial models across 9 platforms.Do free tokens refresh every month?
Not necessarily. OpenRouter and similar platforms do not have a universal “monthly free token refresh” policy. New OpenRouter users receive a small free credit;:free models default to 50 requests per day, rising to 1,000 per day after a $10 top-up. Other platforms calculate free quotas under their own rules, such as Cloudflare’s 10,000 Neurons per day and Groq’s 30 RPM / 1,000 RPD free plan.How do I get free tokens and a free API key?
Register on the relevant platform and create an API key. OpenRouter, Gemini, Groq, Cloudflare, Cerebras, Z.Ai, SiliconFlow, and ZenMux all provide consoles. OpenCode’s free models do not require a separate API key; select them inside the OpenCode client. Most platforms do not require a credit card; Cerebras requires a verified payment method to receive the $5 trial credit.Are the free model APIs OpenAI-compatible?
Most platforms provide OpenAI-compatible endpoints, so you can reuse the OpenAI SDK by changing onlybase_url and api_key. Gemini uses the /v1beta/openai/chat/completions compatibility endpoint, and Cloudflare and Cerebras also offer compatible interfaces. Native API fields may differ slightly on some platforms.Can I call OpenCode's free models directly through the API?
No. The OpenCode free tier is only available inside the OpenCode client (CLI / Desktop / IDE). Callinghttps://opencode.ai/inference/openai/v1/chat/completions directly returns 403 FreeTierError. The free tier also has an undisclosed daily request limit, and LongCat 2.5 has an additional 100 requests/day limit. To call from another client, use a paid model or an OpenCode Go subscription.What usage limits apply to free models?
Common limits include requests per minute/day, context length, concurrency, and feature scope. Inputs and outputs of free models may be used to improve services or train models, so do not submit sensitive information. For production, use paid models or purchase credits.How do I choose the right free LLM?
For the strongest reasoning, choose NVIDIA Nemotron 3 Ultra 550B; for long context, choose Gemini 4 Hydrogen, Gemini 3.8 Flash, Nemotron 3.5 Super VL, or OpenCode’s MiniMax M3.1 Flash / LongCat 2.5 (1M tokens, inside the client); for coding, choose Recursive Flint 1.0, Cohere North Mini Code, Particle-1, OpenCode’s DeepSeek V4.1 Flash Free / GLM-5.3-Flash Free, or Xing4.0 29B; for Chinese conversation, choose GLM-4.7-Flash; for document parsing, choose PaddleOCR-VL 1.5; for translation, choose Hunyuan-MT-7B.What is OpenRouter's free token quota?
New OpenRouter users receive a small free credit. Models marked:free are not billed per token but have daily request limits: 50 requests/day by default, rising to 1,000 requests/day after a $10 top-up. There is no official “monthly refresh” promise; refer to your OpenRouter account page and rate-limit documentation.Usage Example
# Example: call Nemotron 3.5 Super VL
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer ***" \
-H "Content-Type: application/json" \
-d '{
"model": "nvidia/nemotron-3.5-super-vl:free",
"messages": [{"role": "user", "content": "Hello!"}]
}'