Models & model IDs

What a model ID looks like

The canonical ID is maker/name — the same convention our catalogue syncs against:

openai/gpt-4o-mini
anthropic/claude-sonnet-4.5
google/gemini-2.0-flash

Short names also resolve when the operator has defined an alias. If an alias collides with another model's canonical ID, the canonical ID always wins — so a name never resolves to a different model between two requests.

Listing them

The list is always scoped to what that key can actually call — so what you see and what works never disagree.

curl https://iranrouter.com/v1/models -H "Authorization: Bearer ir-..."

Each row carries a billed_from field: 'plan' means it draws on your plan allowance, 'credit' means it bills the wallet.

Each row also carries architecture (what the model takes and makes) and endpoint (the path it answers on): one catalogue holds chat, embedding, image, audio and rerank models, and the id alone does not say which.

The public catalogue with Rial prices lives at the models page, with a dedicated page per model.

Choosing one

• Give simple work (classification, extraction, language detection) to a small model. The price gap between tiers is often an order of magnitude.
• Persian text tokenizes into more tokens than the same text in English, so model choice matters more here than for an English-language team.
• To compare, make one key per model and run the same samples through each — usage reporting then separates the cost for you.

Reaching models outside your plan

If your plan includes only some models and you want a stronger one, you do not have to switch plans and lose what you paid for. Create a key and set its billing source to credit: that key reaches every active model and bills the wallet, while your plan allowance stays untouched.