Every leading AI model. At 25% of the price.

That sounds suspicious, so we're not going to ask you to believe it. Test it for free with credits on us.

Explore models

01 · Your app

base_url "https://glideapi.dev/v1"

02 · Glide routes and meters

glideapi.dev/v1 one key · one endpoint

03 · Official provider APIs

See how it works

Real and official models
Prompts and data never stored
No card required
Refund unused balance anytime

The market today

Right now, every way to buy AI tokens has a downside

Official APIs

OpenAI, Anthropic, Google direct

Demand full pricing forever, along with a separate integration, key and invoice for every provider you add.

Full price

Subscription plans

Claude Max, ChatGPT Plus

Get throttled mid-task. Quotas shrink without warning and you find out when your agent stops halfway through a job.

Throttled

Aggregators

OpenRouter and friends

Give you one endpoint, which is genuinely useful, but it gets you no savings.

No savings

Cheap relays

Telegram sellers, gray-market proxies

Save you money, but with fake models, inflated token counts and stolen data.

Not trustworthy

You're right to be suspicious

Most cheap AI APIs are a trap.

01

Model substitution

They substitute original models with fake ones.

02

Token inflation

They charge you more by inflating tokens, showing more than you actually used.

03

Data logging

They log your prompts, completions and reasoning chains to resell them.

04

Disappearance

They disappear once their temporary exploits die, with no way to contact them.

Introducing Glide API

That's why we give the same models at 25% of the price, and a way to verify them

The model you choose is the model you get

Official models, with the same context windows and the same precision.

Accurate token counts

Because they are the official models. Every response returns input, output, reasoning, and cache creation split by TTL.

We don't store your data

We only keep the timestamp, model, token count, status, cost and balance change. Never your prompts, completions or data.

No risk of losing money

Card payments via Polar (a Stripe partner) and digital assets via OxPay. Unused balance can be refunded on request.

Easy to set up

Switch the base URL in your existing setup and you're running. Nothing else changes.

Low latency, higher uptime

Published live per model. We show live numbers, not a screenshot from a good week.

Don't trust us

Run these five tests with free credits and verify if we are trustworthy

These are the tests that catch a fake AI API, and you can run all of them for free on our credit. Ten minutes, our money.

01 Signature verification

Response · content[0] application/json

{
  "type": "thinking",
  "thinking": "The user just said \"hi\". I'll
              respond naturally and concisely.",
  "signature": "CAISoQUKiAEIDxgCKkAw+kAybZ0hiRUdRRGKnnH2eOft4WT6aMiSTt…"
}

VALID Accepted by the official API on your own key

You validate a signed thinking block from us against the official API on your own key. Only the provider can generate it.

Response · usage application/json

{
  "usage": {
    "input_tokens": 16,
    "output_tokens": 29,
    "output_tokens_details": {
      "thinking_tokens": 20
    },
    "cache_creation_input_tokens": 73,
    "cache_read_input_tokens": 0
  }
}

official reported16 in / 29 out✓
glide returned16 in / 29 out✓ identical
glide charged$0.00038✓ rate × tokens

Three numbers have to agree: what official reported, what we returned, and what we charged for..

PASS · run it yourself Sample output

If any of these fail, don't trust us. Leave.

Get free credits and run the tests Read the full test protocol

Pricing

Compare our live model pricing

Every model is 25% of the official price. Take the published price for any model, multiply by 0.25, and that's what you pay us.

Price per 1M tokens · input / output Loading live rates

Current price per 1M tokens for input and output, compared with the published model-catalog baseline.
Model	Official	Glide API	You save
Loading current model prices...

Prices are shown per 1M tokens. Current cached-input rates and the complete catalog are available on the Models page.

See all models

Calculate your own savings

Live · current catalog rates

Model

Monthly input tokens

40M

Monthly output tokens

8M

Monthly readout USD

What you're paying now: $0.00
What you'd pay us: $0.00
You save every month Live: $0.00
Over a year: $0.00

Figures use current rates returned by the public model catalog. Start with free credits

Before you spend anything

Still thinking? We are risk-free and secure

Free Credit

On sign up

Enough to run all five tests above and still have credit left to use. Works on every model in the catalog. No card required.

Refund any unused balance, any time

If there's unused credit in your account and you want it back, just ask and we will send it without any conditions.

Prepaid, never a subscription

No minimum threshold for a recharge, you can even top-up a dollar. Nothing renews or auto-charges.

Payments that clear properly

Cards through Polar (a Stripe partner), digital assets through OxPay. No wallet addresses in DMs, no shady Telegram messages.

A support channel with humans behind it

Email or Discord. We usually reply within 2–5 hours (because we're humans), and we never leave you hanging.

Zero Data Retention

ZDR by default

Your prompts, responses, tool payloads, and API keys are never stored, logged, or used for training. Requests are relayed and forgotten — there is no conversation history on our side to breach, sell, or subpoena.

Setup

It's easy to set up. Just change the base URL.

Step 1

Get a key

Create an account and generate a key in the console.

Step 2

Point at us

Replace one line in your existing client. Pick your stack in the terminal below.

Step 3

Pick a model

Only the model string changes. Pull current IDs from GET /v1/modelsFull endpoint reference and per-client setup.

~/your-project · point at glideapi.dev

python · openai

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_GLIDE_KEY",
    base_url="https://glideapi.dev/v1",
)

javascript · openai

const client = new OpenAI({
  apiKey: process.env.GLIDE_API_KEY,
  baseURL: "https://glideapi.dev/v1",
});

python · anthropic

from anthropic import Anthropic

client = Anthropic(
    api_key="YOUR_GLIDE_KEY",
    base_url="https://glideapi.dev",
)

bash · claude code

export ANTHROPIC_BASE_URL="https://glideapi.dev"
export ANTHROPIC_AUTH_TOKEN="YOUR_GLIDE_KEY"
export ANTHROPIC_MODEL="YOUR_MODEL_ID"
claude

toml · codex cli · config.toml

model_provider = "glideapi"
model = "YOUR_MODEL_ID"

[model_providers.glideapi]
name = "GlideAPI"
base_url = "https://glideapi.dev/backend-api/codex"
env_key = "GLIDE_API_KEY"
wire_api = "responses"
requires_openai_auth = false

Base URL is the only change. Keys, payloads and responses stay the same shape.

Also works with

LangChainCline Roo CodeContinueAider OpenCode n8nDifyFlowise Open WebUILibreChatLobeChat AnythingLLM

Read the Docs

FAQ

Questions worth asking

More questions? Email us or join the Discord.

What stops you from selling my prompts or data?

We don't have them. Requests are routed, not retained. Your prompts and completions aren't written to our storage, so there's nothing to sell, leak or hand over.

What we do keep is the billing record: timestamp, model, token count, status, cost, balance change.

Every cheap provider says their models are "verified." Why are you different?

We're not different because we say it louder. We're different because we published the tests that would expose us, and told you to post the results if we fail.

Scroll back to the five tests. That protocol works on any provider. Run it on us, then run it on whoever else you're considering, and notice which of us published it.

Are the models quantized?

No. Full precision. No FP8, no INT4, no "optimized" variants. Test 2 catches quantization if you don't take our word for it. Quantized weights diverge from official output at temperature zero, quickly and visibly.

What happens to my balance if I stop using you?

You ask for it back and we send it. Any unused balance, any time, no conditions and no expiry window. It's prepaid, not committed.

Do I get throttled like on a subscription?

No. There's no seat, no weekly quota and no usage cap that resets on someone else's schedule. You bought tokens, you spend tokens.

That's the difference this page exists to sell. A subscription rations you. A balance doesn't.

What if you go down?

Per-model uptime, latency and status are published live at status.glideapi.dev. Current numbers, not a screenshot from a good week. When a route degrades we fail over to another authorized route to the same model; if none is available you get an error rather than a substitute.

If something breaks, is there anyone to talk to?

Yes, and you can check that before you sign up. Email or Discord, with an average first response under two hours.

Does prompt caching work?

Yes. Cached input is priced separately when the selected model supports it. Check the live Models catalog for the current cached-input rate.

Can I see what each request actually cost me?

Every one of them, listed. Input tokens, output tokens, model, status and cost, per request, from your first call, in the dashboard and through the API:

/v1/usage/logs line items · /v1/usage/models spend by model · /v1/usage/keys spend by key · /v1/balance what's left

Don't trust us.Test us.

Take the free credits. Run the tests. Check every number we've claimed on this page against the official API.

If we're lying, you'll know in ten minutes and it cost you nothing. If we're not, you just cut your AI bill by 75%.

Sign up and claim free credits Talk with us