OpenAI-compatible API access, paid in tokens.

Create a key, redeem a battery code into it, and point any OpenAI client at the base URL. No account needed: your API key is your login.

1
Create or import a keyIt stays on this browser until you clear it. The secret and the name are kept here, never on the server.
Add a key
A new key starts empty. Redeem a battery code into it to add tokens.
Name (optional)
2
Redeem battery codesA BAT-… code adds tokens to the key it is redeemed into. Its validity starts at redemption.
3
Point your client at the APICodex, Cursor, the OpenAI SDK, curl: anything that speaks the OpenAI Chat Completions API.

Integration

OpenAI-compatible · streaming supported · only whitelisted models are served

API base URL

https://api.aicredits.cc/v1

POST/v1/chat/completions

GET/v1/models

GET/v1/pricing

AUTHAuthorization: Bearer sk-…

Codex CLI defaults to the Responses API; this gateway speaks Chat Completions, so keep wire_api = "chat". Rotate a key instead of creating a new one to keep its batteries.

config.toml

# ~/.codex/config.toml
model = "gpt-5.6-sol"
model_provider = "aicredits"

[model_providers.aicredits]
name = "aicredits"
base_url = "https://api.aicredits.cc/v1"
env_key = "AICREDITS_API_KEY"
wire_api = "chat"

# then:  export AICREDITS_API_KEY=sk-...   &&  codex

Multipliers

What a request costs, per model. 1 token = the cost of 1 input token on gpt-5.6-sol. Pricing v2.

Model	Input	Cached input	Output	1M tokens buys
GPT-6 Astragpt-6-astra	2×$10.00 / 1M	0.2×$1.00 / 1M	10×$50.00 / 1M	500,000 input100,000 output
GPT-5.6 Solbasegpt-5.6-sol	1×$5.00 / 1M	0.1×$0.50 / 1M	6×$30.00 / 1M	1,000,000 input166,667 output
GPT-5.6 Terragpt-5.6-terra	0.4×$2.00 / 1M	0.04×$0.20 / 1M	2.4×$12.00 / 1M	2,500,000 input416,667 output
GPT-5.6 Lunagpt-5.6-luna	0.04×$0.20 / 1M	0×$0.02 / 1M	0.24×$1.20 / 1M	25,000,000 input4,166,667 output

A request costsceil( input × k_in + cached × k_cache + output × k_out )

For example

1,000 input and 200 output tokens on gpt-5.6-sol cost 2,200 tokens.

Good to know

Reasoning counts as output, as the provider bills it. With no usage reported the charge is estimated and rounded down. Price changes ship as a new version with an effective date, never retroactively.