Skip to content
GPT-6 Astra is now available on every paid plan.See it

Every frontier model, on one key.

An OpenAI-compatible endpoint in front of Claude, GPT, Gemini, Grok and MiniMax. Go is $5 a month for $33–$86 of API credit — a genuinely unfair deal.

Get $33–$86 of creditBrowse 19 models
terminal
export OPENAI_BASE_URL="https://ai.eaon.dev/v1"
export OPENAI_API_KEY="eaon_sk_..."

# GPT-6 Astra, Opus 5, and 17 more answer here
models
19
models
providers
5
providers
API credit for $5/mo
$33–$86
API credit for $5/mo
max context
1M
max context

Models from

OpenAIAnthropicGooglexAIMiniMax

Where most people start

All 19 models

Buying model access should be boring

  • One bill instead of five

    Anthropic, OpenAI, Google and xAI each want their own account, their own key and their own invoice. Here it is one line on one card.

  • The deal is the product

    Go is $5 a month for $33–$86 of API credit. Plus is $25 for $165–$430. Same models as the official APIs, metered the same way — just a fraction of the bill.

  • One flat fee, no rate limits

    A single subscription instead of a card on file per token. Every call spends real tokens against that credit. Nothing here throttles how fast you can call it.

  • Nothing to migrate to

    The endpoint is OpenAI-shaped. Change a base URL and a key in whatever you already use, and the models change underneath you.

What a plan actually buys

Pick a plan to see roughly how far its usage grant goes on each model, or filter by tier. Every call is metered from real tokens — no private multipliers.

Plus: $165–$430 of API credit a month for $25. That’s $41.25–$107.50 each week (monthly ÷ 4) — unused weeks do not roll over. Every call is metered from the tokens it actually used.

ModelContextTierOn Plus
MiniMax M2.5eaon/minimax-m2.5200kflashIncluded
MiniMax M2.5 HighSpeedeaon/minimax-m2.5-highspeed200kflashIncluded
MiniMax M2.7eaon/minimax-m2.7200kflashIncluded
MiniMax M2.7 HighSpeedeaon/minimax-m2.7-highspeed200kflashIncluded
MiniMax M3eaon/minimax-m3200kflashIncluded
Gemini 3.1 Flash Liteeaon/gemini-3.1-flash-lite1,000kflashIncluded
Gemini 3 Flasheaon/gemini-3-flash1,000kflashIncluded
Gemini 3.7 Flasheaon/gemini-3.7-flash1,000kflashIncluded
Gemini 3.8 Flasheaon/gemini-3.8-flash1,000kflashIncluded
Grok 4.5eaon/grok-4.5256kstandardIncluded
Grok 4.6eaon/grok-4.6256kstandardIncluded
GPT-5.6 Terraeaon/gpt-5.6-terra400kstandardIncluded
Claude Sonnet 5eaon/sonnet-51,000kstandardIncluded
GPT-5.5eaon/gpt-5.5400kstandardIncluded
GPT-5.6 Soleaon/gpt-5.6-sol400kfrontierIncluded
Gemini 3.1 Pro Previeweaon/gemini-3.1-pro-preview1,000kstandardIncluded
Claude Opus 5eaon/opus-51,000kfrontierIncluded
GPT-6 Astraeaon/gpt-6-astra400kfrontierIncluded
Claude Fable 5eaon/fable-5200kfrontierIncluded

How it works

Three steps, and none of them involve talking to anyone.

  1. 01

    Create a key

    Sign up and issue one from the dashboard. It is shown once, and it works against every model your plan covers.

  2. 02

    Point your client at it

    Set the base URL to the Eaon endpoint and the key to your new one. No SDK to install and no proxy to run locally.

  3. 03

    Pick a model per call

    Send an Eaon model id in the request body. Swap it whenever you like — each call is metered from the tokens it actually used.

Drop it into what you already use

Two environment variables. No SDK to install, no wrapper library, no proxy running on your machine.

~/.zshrc
export ANTHROPIC_BASE_URL="http://localhost:3000/v1"
export ANTHROPIC_AUTH_TOKEN="eaon_sk_..."
export ANTHROPIC_MODEL="eaon/sonnet-5"

claude

Four things you can check yourself

  • We cannot show you a key twice

    Only a sha256 of each key is stored. The plaintext appears once, when you create it, and after that we do not have it either. Lose one and you revoke it and issue another.

  • An outage on their side does not spend your month

    Budget is reserved before the upstream call and handed back if it fails.

  • Cancelling happens on the page you subscribed from

    Billing runs through Stripe end to end. No email to write, no retention flow to sit through.

  • The status page is allowed to say we are broken

    It reports success rate and latency measured from the request log. When a model has not been called enough to judge, it says that instead of showing a number built on four requests.

See what the gateway is doing right now

Three plans, and you can change your mind

Start free, move up when you hit the ceiling, cancel from the same page you subscribed on.

Free

$0/month

Enough to wire it up and see it work.

  • About $3.75–$8.50 of API-equivalent value
  • Resets monthly (not weekly)
  • Flash models
  • 1 API key
  • No rate limits

Go1× Go

$5/month

$33–$86 of API credit for $5.

  • $33–$86 of API credit a month
  • ≈$8.25–$21.50 each week (monthly ÷ 4)
  • Unused week does not roll over
  • Every model, GPT-6 Astra included
  • 5 API keys
  • No rate limits

Monthly credit is billed in 7-day windows: you get 1/4 each week. Meter resets every week from your plan start date.

Plus5× Go

$25/month

5× Go — $165–$430 of API credit.

  • $165–$430 of API credit a month
  • ≈$41.25–$107.50 each week (monthly ÷ 4)
  • Unused week does not roll over
  • Every model, GPT-6 Astra included
  • 10 API keys
  • No rate limits

Monthly credit is billed in 7-day windows: you get 1/4 each week. Meter resets every week from your plan start date.

Pro10× Go

$50/month

10× Go — $330–$860 of API credit.

  • $330–$860 of API credit a month
  • ≈$82.50–$215 each week (monthly ÷ 4)
  • Unused week does not roll over
  • Every model, GPT-6 Astra included
  • 10 API keys
  • No rate limits

Monthly credit is billed in 7-day windows: you get 1/4 each week. Meter resets every week from your plan start date.

Max20× Go

$100/month

20× Go — $660–$1,720 of API credit.

  • $660–$1,720 of API credit a month
  • ≈$165–$430 each week (monthly ÷ 4)
  • Unused week does not roll over
  • Every model, GPT-6 Astra included
  • 20 API keys
  • No rate limits

Monthly credit is billed in 7-day windows: you get 1/4 each week. Meter resets every week from your plan start date.

Questions that come up

What am I actually billed for?

The real cost of the call: prompt tokens, reply tokens, and cached tokens at a lower rate, priced per model. A short question and a long conversation on the same model cost different amounts, honestly, rather than both being rounded up to one flat charge.

Does unused budget roll over?

No. It resets every 7 days from when your plan started (every 30 on Free). We priced the plans expecting most windows to leave some on the table, which is what lets the heavy ones stay affordable.

Are there rate limits?

No. Send requests as fast as you want — nothing here throttles a resolved key. The only limit that applies is your account's own budget, and a spike just spends it faster.

Will my existing tools work without changes?

If they speak the OpenAI chat completions protocol, yes. Point the base URL at Eaon, swap the key, and pick a model id from the catalogue. Claude Code, Cursor, Codex CLI and the official SDKs all work this way.

What happens when I run out?

A request whose worst case would exceed what's left gets HTTP 402 immediately, before it ever reaches a model — so you're never charged for a call that didn't run. Nothing silently downgrades to a cheaper model behind your back either.

Do you train on my requests?

No. Prompts pass through to the upstream provider and are not stored. What we keep is the metering row: model, real cost, token counts, latency and status.

Can I cancel whenever?

Yes, from the billing portal. The plan stays live until the end of the period you already paid for, then drops to Free.

New models, when we add them

Rate changes too. Nothing else, and never more than once a month.