How to Get Your Replicate API Key (100k+ Open Source Models)
Every open-source model ever published, through a single API

Replicate gives you access to virtually every open-source model ever published, via a single API. If it exists as an AI model, it's probably on Replicate. One token covers LLMs, image generation, video, audio, and hundreds of specialized tools.

Prerequisites

• A replicate.com account (GitHub SSO available - fastest way to sign up)
• No credit card required to start - some models offer free predictions

Get your token in 3 steps

1. Sign up and log in

Go to replicate.com and create an account. GitHub single sign-on is available and makes the process instant.

2. Navigate to "API tokens"

Click your avatar in the top right corner. Select API tokens from the dropdown menu.

3. Copy your token

You'll see a default token already created for your account. Click it to copy, or click Create token to generate a named token for a specific project. Store it securely - it won't be shown again in full if you create a new one.

Models unlocked with your token

Replicate hosts over 100,000 community and official models. Here's a sample of what's available:

LLMs:

• Llama 4 and Llama 3 series - Meta's open models with instruction tuning
• Mistral variants - including fine-tunes and specialized versions
• Hundreds of community fine-tunes for specific domains and languages

Image generation:

• Flux (all variants) - Schnell, Dev, Pro, and community fine-tunes
• SDXL and dozens of specialized SDXL checkpoints
• Specialized models for portraits, illustrations, product photos, and more

Video:

• CogVideoX - high-quality text-to-video generation
• AnimateDiff - animation from still images
• Many more video models from the open-source community

Audio:

• MusicGen - text-to-music generation
• AudioCraft - audio generation and effects
• Voice cloning, speech synthesis, and transcription models

Tools:

• Background removal, image upscaling, face restoration
• ControlNet variants for guided generation
• Inpainting, outpainting, style transfer

If a model was published as open source in the last few years, there's a high probability it's already running on Replicate.

Pricing

Replicate charges per prediction, with pricing that varies by model:

• Many small models cost fractions of a cent per run
• Community models often have free tiers for low-volume use
• Popular models like Flux and Llama are priced competitively against dedicated providers
• You pay only for compute time, no base subscription required

Check the model page for exact pricing before running at scale - costs vary significantly between a 7B LLM and a video generation model.

Security tips

• Never hardcode your token in source code or commit it to a Git repository
• Use environment variables (.env files) in all your projects
• Create named tokens per project so you can revoke one without breaking everything
• Monitor usage in the Replicate dashboard - unusual spikes can indicate a leaked token

Why Replicate for model access

Self-hosting 100,000 models is not a realistic option for any team. Replicate solves the cold-start problem, handles model versioning, and provides a consistent REST API across every model. The same code pattern that calls a Flux image model can call a CogVideoX video model - only the model identifier changes.

For teams that need to experiment across model categories without infrastructure commitment, Replicate is unmatched in breadth.

Connect your token to HiWay2LLM

Once you have your Replicate token, bring it to HiWay2LLM in seconds. Route your Replicate calls through a unified endpoint with cost tracking, usage analytics, and full observability built in.