Choosing the model and understanding costs

What it is

Every provider offers several models, from cheap-and-fast to powerful-and-pricier. The model you pick decides both the quality of the generated SEO text and what you pay. In bring-your-own-key mode you are billed by the provider in tokens (roughly: pieces of words in and out); on AISA Cloud you are billed in credits instead, where 1 credit = 1 optimised content item. This article explains how to choose a model in AISA and what each one costs.

Where to find it

AISA → Settings → API Keys. Each provider card has a model dropdown with a short cost/quality note per model, and a separate vision-model dropdown for providers that support images. Estimated costs also appear before you run a batch in 1-Click SEO and Bulk SEO (in tokens for BYOK, in credits for Cloud).

Requirements & tier

Model choice applies on Free and Premium (BYOK) tiers. On Cloud the server selects the model for you and bills credits.
One configured, enabled provider with a valid key.
Note that DeepSeek tokenises differently; AISA uses a 1024-token base budget for text, scaled up for longer jobs like Q&A and llms.txt.

How to use

Open the provider card in Settings → API Keys and read the cost note (each model line shows whether it is the cheap, balanced or most-powerful option).
For everyday SEO meta, pick the provider's cheapest model — for example DeepSeek-chat, GPT-4o Mini, Gemini 2.5 Flash, Claude Haiku 4.5, or Mistral Small. Meta titles and descriptions are short, so the cheap models are usually enough.
Choose a stronger model (GPT-4o, Sonnet 4.6/Opus 4.8, Gemini 2.5 Pro, Mistral Large) only when you want richer long-form output such as detailed AI Context or FAQ sets.
If your provider supports Vision, set the image model separately in the "Images & Vision" box (a cheaper vision model is usually fine for alt-text).
Before a big run, read the cost/credit estimate shown in 1-Click SEO or Bulk SEO so there are no surprises.
Save and generate.

Questions people ask

What does a typical optimisation cost? With DeepSeek about €0.0002 per content item; OpenAI GPT-4o Mini and Gemini 2.5 Flash are in a similar range; GPT-4o or Claude Sonnet cost a few thousandths of a euro each.
Does the 1-Click batch save money? Yes. One-Click sends a single API call that returns all the SEO for a page as one JSON, instead of separate calls — roughly 60% fewer tokens. It is available on Pro/Agency/Studio and Cloud (Free uses separate calls).
Tokens or credits — which do I count? Free and Premium count tokens (you pay the provider). Cloud counts credits (1 credit = 1 optimised item). The UI follows whichever mode you are in.
Is a more expensive model worth it for meta tags? Rarely. SEO titles/descriptions are short and constrained; the cheap models already meet the on-page rules. Save the big models for long content.
Which Gemini model should I choose? gemini-2.5-flash is the recommended default (2.5 Flash-Lite is even cheaper). Avoid old 2.0-flash — it was deprecated.
Can I use a different model for text and for images? Yes. The text model and the vision model are set independently per provider.

Troubleshooting

If output looks truncated on long jobs (llms.txt, big FAQ sets), it is the token budget — longer tasks already get a larger budget, but very long content can still clip; split it or use a model with a bigger context window (for example DeepSeek V4 Pro's 1M context). If costs feel high, switch the everyday text model down to the cheapest tier and reserve powerful models for long-form. If a chosen model errors as unavailable, it may have been deprecated by the provider — pick a current model from the dropdown (for Gemini, the "always latest" Flash option is deprecation-proof).

Supported AI providers and which one to choose
Connect an AI provider and add your API key
Bring your own key (BYOK) vs AISA Cloud credits
AISA Turbo: parallel speed and auto-tuning on rate limits