Models and reliability
How Aimogen Pro chooses a model, what capability lists control, and the reliability machinery that keeps generation working when a provider fails.
Aimogen Pro exposes model choice almost everywhere: per rule, per chatbot, per agent, per OmniBlock, per shortcode. This section explains how that choice is made, what happens when it fails, and how to keep costs predictable.
In this section#
The short version#
Model choice is per feature, not global. Every model selector lists every model from every configured provider, annotated with its provider. Routing is derived from the model name at request time. See AI provider configuration.
Capability lists gate features. The plugin keeps internal lists of which models support vision, tool calling, streaming, embeddings and so on. A selector only offers models that can do the job.
Failures are classified, not just retried. Errors are grouped into safety, context, quota, authentication, permission, model, rate limit, network and provider categories, and each category is handled differently.
Reliability Mode is off by default. With it off you get the legacy behaviour: a fixed retry count and a randomly chosen fallback model. With it on you get ordered fallbacks, capability-aware skipping, cooldowns and a provider circuit breaker.
The default model#
AIMOGEN_DEFAULT_MODEL is gpt-5-mini in 2.8.8. It applies when a feature has no explicit model set.
Settings › General Settings › Default AI Model To Use When No AI Model Available overrides it. Set this to a model from a provider you have actually configured — otherwise features with no explicit model will fail on a site that never configured OpenAI.
Global generation parameters#
These live on Settings › AI Writer and apply to the writer unless a feature overrides them.
| Setting | Default | What it controls |
|---|---|---|
| Temperature | 1 | Randomness. Lower is more predictable and repetitive, higher is more varied and more likely to wander |
| Top_p | 1 | Nucleus sampling. Adjust this or temperature, not both |
| Presence penalty | 0 | Discourages reusing topics already mentioned. Positive values push toward new subject matter |
| Frequency penalty | 0 | Discourages repeating the same words. Useful against repetitive phrasing in long articles |
| Max tokens | 2048 (AIMOGEN_DEFAULT_MAX_TOKENS) | Cap on generated output. Bulk content uses 4096 |
The chatbot, the editor, OmniBlocks and each rule have their own copies of these, so you can be conservative globally and adventurous in one place.
Three filters let code override them per request: aiomatic_temperature, aiomatic_top_p,
aiomatic_presence_penalty and aiomatic_frequency_penalty. See
Filter reference.
GPT-5 specific controls#
Two settings on Settings › Advanced AI Settings apply to GPT-5 family models:
| Setting | Values | Effect |
|---|---|---|
| Reasoning Settings | empty, minimal, low, medium, high | How much internal reasoning the model does before answering. Higher costs more tokens and more time |
| Verbosity Settings | empty, low, medium, high | How expansive the answer is |
Leave both empty to use the provider defaults. Raise reasoning for analysis and structured extraction; leave it low for bulk article writing, where it mostly buys latency.
Deterministic output#
Settings › Advanced AI Settings › Seed For AI Writer sets the sampling seed. With the same seed, the same prompt and the same parameters, a provider makes a best effort to return the same output.
The plugin help text is careful here, and correctly so: determinism is not guaranteed, because providers change model backends over time. Use a seed to reduce variance during prompt development, not as a guarantee.
Related#
- AI provider configuration
- Usage logs and statistics — what each request actually cost
Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.