OpenAI provides the GPT family and o-series reasoning models. Access via API key or through OpenAI Codex OAuth.
Setup
Get your API key from the OpenAI Platform.
export OPENAI_API_KEY=sk-...
Config
{
"providers": {
"openai": "${OPENAI_API_KEY}"
}
}
Use it
{
"agents": [
{ "name": "coder", "model": "openai:gpt-4o" }
]
}
openai from gpt-, o1-, o3-, o4-, chatgpt-, and codex- prefixes.
Models
GPT-5 Series
| Model | Best for | Context | Reasoning |
|---|
gpt-5.2 | Latest flagship — fast reasoning, strong coding | — | Yes |
gpt-5.1 | Great for coding and agentic tasks | — | Yes |
gpt-5 | Flagship general-purpose | — | Yes |
gpt-5-mini | Smaller 5-series, cost-effective | — | Yes |
GPT-4 Series
| Model | Best for | Context | Reasoning |
|---|
gpt-4o | Multimodal — text, vision, audio | 128K | No |
gpt-4.1 | Improved instruction following | 1M | No |
gpt-4o-mini | Cheap and fast for simple tasks | 128K | No |
O-Series (Reasoning)
| Model | Best for | Context | Reasoning |
|---|
o3 | Deep reasoning, complex logic and math | 200K | Yes |
o3-pro | More compute for harder problems | 200K | Yes |
o4-mini | Fast reasoning at lower cost | 200K | Yes |
o3-mini | Small reasoning model | 200K | Yes |
gpt-5-nano, gpt-5-pro, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.2-codex, gpt-4.1-mini, gpt-4.1-nano, codex-mini-latest
The o-series models (o3, o4-mini) use chain-of-thought reasoning. They’re slower but significantly better at complex logic, math, and multi-step planning. Good choice for the orchestrator model or a reviewer agent.
Features
| Feature | Supported |
|---|
| Streaming | Yes |
| Tool use | Yes |
| Vision (images) | Yes (GPT-4o, GPT-4.1, GPT-5.x) |
| Reasoning | Yes (o-series, GPT-5.x) |
Pricing
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|
gpt-5.2 | $1.75 | $14.00 |
gpt-5 | $2.00 | $8.00 |
gpt-4o | $2.50 | $10.00 |
gpt-4.1 | $2.00 | $8.00 |
o3 | $10.00 | $40.00 |
o4-mini | $1.10 | $4.40 |
gpt-4o-mini | $0.15 | $0.60 |
Provider Details
| |
|---|
| Provider ID | openai |
| Env variable | OPENAI_API_KEY |
| API type | OpenAI Responses API |
| Auto-infer prefixes | gpt-, o1-, o3-, o4-, chatgpt-, codex- |
| Base URL override | Supported |
OAuth: Use Your ChatGPT Subscription (Codex)
If you have a ChatGPT Plus or Pro subscription, you can use it directly via the OpenAI Codex OAuth flow — no separate API key or billing account needed.
Supported Subscriptions
| Plan | Works | Notes |
|---|
| ChatGPT Pro | Yes | Highest rate limits |
| ChatGPT Plus | Yes | Standard rate limits |
| ChatGPT Free | No | Insufficient API access |
Login
polpo auth login openai-codex
- Polpo opens a browser window for authorization
- Log in with your OpenAI/ChatGPT account
- The callback server on
localhost:1455 captures the token automatically
- If the callback fails, you can paste the redirect URL manually
Logging in to OpenAI Codex (ChatGPT Plus/Pro)...
Opening browser for authorization...
Waiting for callback on localhost:1455...
Login successful!
Profile saved as: openai-codex:user@gmail.com
Stored in ~/.polpo/auth-profiles.json
Port 1455 must be available on your machine. If another service is using it, the callback will fail — in that case, paste the redirect URL manually when prompted.
Provider ID
When using Codex OAuth, the provider ID is openai-codex (not openai):
{
"agents": [
{ "name": "coder", "model": "openai-codex:gpt-4o" }
]
}
openai (API key) and openai-codex (OAuth). Polpo resolves whichever has a valid credential.
Token Refresh
OAuth tokens expire. Polpo handles this automatically:
- Before each API call, Polpo checks if the token is expired
- If expired and a refresh token exists, it refreshes silently
- If no refresh token is available, the profile is skipped and Polpo falls back to other credentials
Multiple Accounts & Profile Rotation
You can log in with multiple OpenAI accounts. Each gets its own profile:
openai-codex:user@personal.com
openai-codex:user@work.com
lastUsed first, skipping profiles in cooldown or billing disable.
Billing Disable
If your ChatGPT subscription quota is exceeded, the profile enters billing disable — a longer backoff (5h → 10h → 20h → 24h cap) than regular rate-limit cooldown. Other Codex profiles or openai API keys continue working normally.
Session Pinning
Pin a specific Codex profile for the session:
/model openai-codex/gpt-4o@openai-codex:user@work.com
/model reset.
Resolution Order
1. polpo.json providers.openai.apiKey (config wins)
2. OPENAI_API_KEY environment variable (env var)
3. Stored openai-codex OAuth profiles (auto-refresh)
For the full authentication system — all OAuth providers, CLI commands, profile rotation algorithm, and security details — see the Authentication guide. Notes
- The GPT-5 series represents OpenAI’s latest generation with built-in reasoning and vision.
- GPT-4.1 has a 1M token context window — useful for large codebase tasks.
- Use
gpt-4o-mini for the “fast-worker” pattern: assign it formatting, boilerplate, and simple refactoring where speed and cost matter more than peak quality.
- o3 is the strongest standalone reasoning model from OpenAI — consider it for reviewer agents and complex planning tasks.
- Codex variants (
gpt-5.1-codex, gpt-5.2-codex) are optimized for long-running agentic coding tasks.
