Skip to main content
Bytez uses a credit-based system. Credits are consumed when you run models, and how they’re consumed depends on whether you’re using closed-source or open-source models.

​
Plans

Free

  • Run open models up to 7B parameters
  • Access all closed model providers
  • 1 concurrent request (open models)
  • 10 requests/second (closed models)
  • Credits refresh every 4 weeks

Pay-as-you-go

  • Run open models up to 120B parameters
  • Access all closed model providers
  • Rate limits scale with credits purchased
  • Unlimited closed model requests
  • Add credits anytime

​
How Credits Work

Credits are a unified currency across all models on Bytez. When you run a model, credits are deducted from your balance based on usage.
Model TypeHow Credits Are Consumed
Closed modelsBased on provider pricing (per token, per image, per video, etc.)
Open modelsPer second of inference
Your credits purchased in the last 4 weeks determine two things:
  1. Which open models you can access - Larger open models require more credits purchased to unlock
  2. Your rate limits - More credits purchased unlocks more concurrent requests
Adding credits immediately unlocks higher tiers. You don’t need to wait for the next billing cycle.

​
Credit Unlock Thresholds

Credits Purchased (last 4 weeks)Open Model AccessConcurrent Requests (7B)
$0 (Free)Up to 7B1
$3+Up to 7B4
$10+Up to 35B4
$25+Up to 70B10
$50+Up to 120B20
$100+Up to 120B40
$500+Up to 120B200
$1,000+Up to 120B400
Credits expire 4 weeks after purchase. Use them or lose them!

​
Closed Model Billing

For closed-source models (OpenAI, Anthropic, Google, Mistral, Cohere), we pass through the provider’s pricing plus a small platform fee. 
Your cost = Provider price + 2% platform fee
Providers charge differently depending on the model and modality - per token for text, per image for image generation, per second for video, etc. We pass through whatever the provider charges. Example: If OpenAI charges per M tokens, you pay per M tokens.
The platform fee covers:
  • Unified API translation and standardization
  • Request routing and load balancing
  • Usage tracking and analytics
  • Support and reliability infrastructure
You get a single API, single billing, and single format across all providers.

​
What’s included

  • Pass-through pricing - Pay only for what the provider charges
  • No minimum - No monthly minimums or commitments
  • Real-time pricing - We pass through provider rates as they change

​
Open Model Billing

Open-source models run on our serverless GPU infrastructure. You’re billed per second of inference time - no cold start fees, no idle charges. 
Your cost = Inference time (seconds) x Rate for model size

​
Pricing by Model Size

Bigger models use more VRAM, so they cost more per second:
Model SizePer SecondPer Hour
7B$0.000072~$0.26
15B$0.000108~$0.39
35B$0.000144~$0.52
70B$0.000216~$0.78
120B$0.00036~$1.30
Our base rate is $0.0000045/GB-second of VRAM used.For comparison:
  • Bytez: $0.0000045/GB-sec (with Nvidia GPUs)
  • AWS Lambda: $0.0000167/GB-sec (CPUs only)
That’s 3.7x cheaper than AWS Lambda, and you get serverless Nvidia GPUs, not just serverless CPUs.

​
What’s included

  • Per-second billing - Billed in 1-second increments
  • No cold start fees - You don’t pay while the model loads
  • No idle charges - You don’t pay when not running inference
  • No reserved instances - No commitments, no minimums

​
Auto-Reload

Auto-reload automatically tops up your credit balance when it runs low, so your API calls never fail unexpectedly.

​
How it works

SettingDefaultDescription
Threshold$3Reload triggers when balance drops below this
Reload amount$10Amount added to your balance
Monthly max$100Maximum auto-reload spend per month
1

Balance drops below threshold

When your credit balance falls below $3 (default), auto-reload activates
2

Card is charged

Your saved payment method is charged $10 (default reload amount)
3

Credits are added

$10 in credits is immediately added to your balance
4

Monthly cap enforced

Auto-reload stops if you’ve hit your monthly maximum ($100 default)

​
If Auto-Reload is Disabled

When auto-reload is off and your credits run out, you may get an API response like this: 
{
  "status": 402,
  "error": "Payment Required",
  "message": "Insufficient credits. Please add credits to continue."
}
If you’re running production workloads, we recommend enabling auto-reload to prevent unexpected failures.

​
Configuring Auto-Reload

You can enable, disable, or adjust auto-reload settings in your API Dashboard.

​
Auto-Scaling (Open Models)

By default, if you exceed your open model rate limits, requests are rejected with a rate-limit error. If you want your rate limits to automatically scale with your traffic in production, add autoScale: true to your request: 
const response = await fetch('https://api.bytez.com/v1/chat/completions', {
  method: 'POST',
  headers: {
    'Authorization': API_KEY,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    model: 'meta-llama/Llama-3-70b',
    messages: [...],
    autoScale: true
  })
});
When enabled, the system auto-purchases extra credits required to keep auto-scaling. You can control your Max Monthly Spend in your API Dashboard to cap costs. This way you can auto-scale and control your budget.
For closed models, you get unlimited rate limits on a pay-as-you-go basis - no auto-scaling needed.

​
Billing Cycle

  • Billing: None
  • Credits: free credits, refreshed every 4 weeks
  • Expiration: Credits expire 4 weeks after grant
  • Billing: charged on signup date
  • Credits: $5 in credits granted each billing cycle
  • Expiration: All credits expire 4 weeks after grant

​
Adding Credits Mid-Cycle

You can add credits at any time. When you do:
  1. Immediate access - Higher model tiers and rate limits unlock instantly
  2. No proration - You get the full credit amount immediately
  3. Credits stack - Purchased credits add to your existing balance
Example: You’re on Pay-as-you-go with remaining. You add . Your new balance is , which immediately unlocks 70B models and 10 concurrent requests.

​
FAQ

In-flight requests will complete. Only new requests will fail with a 402 error.
Credits are non-refundable and expire 4 weeks after purchase.
Visit your API Dashboard to see real-time usage, credit balance, and request history.
Larger models require more GPU resources (VRAM). Requiring a minimum purchase threshold ensures you have enough credits to complete meaningful workloads without running out mid-task.
For high-volume usage (>/month), contact us at [email protected] for custom pricing.