Choosing the AI model for a filter
Each Tender Navigator filter has an LLM model setting that selects which large language model evaluates the tenders it picks up. The list offers several general-purpose models from third-party providers; the default is the first in the list, and a filter keeps working on a retired model until you change it.
Last updated
The setting
In the filter editor, under AI evaluation, the LLM model field chooses the AI model that will assess tender relevance based on your prompt. The editor notes that models differ in depth of reasoning, response speed and evaluation accuracy. The setting is per filter, so two filters can use different models.
The models on offer
As of this article's last update, the list contains:
| Model | Note |
|---|---|
| GPT-5.4 Mini | The default for a new filter |
| MiMo 2.5 | |
| DeepSeek V4 Flash | |
| Gemini 3 Flash (Preview) | A preview model; behaviour may change |
| Qwen 3.5 | |
| Mistral Medium 3.5 |
These are general-purpose models from third-party providers. Tender Navigator may change the list to maintain quality, security and legal compliance; the current provider list is available on request, as the AI Transparency Notice explains.
Retired models
When a model is retired from the list, filters that still use it keep working and the editor shows the model with a "(legacy)" mark. Nothing changes until you pick another model and save. Retired models seen in the editor include Gemini 2.0 Flash, GPT-5 Mini, Grok 3 Mini, MiMo v2 Flash and DeepSeek v3.2.
What changing the model does
- Future evaluations by the filter use the new model from the next daily run.
- Existing results are not re-evaluated. Each result records the model it was evaluated with, visible under Filter info in the Tender Info panel.
Choosing
There is no single best model for every prompt. A practical approach: keep the default, read the reasoning on the first week of results, and try another model on a duplicate of the filter if the reasoning is consistently shallow or slow to arrive. The prompt usually matters more than the model; see Writing the AI prompt.
Related articles
- How AI tender evaluation works in Tender NavigatorHow Tender Navigator's AI evaluation judges each tender a filter picks up against the filter's prompt, what the model receives, what it returns (Matched or Unmatched with reasoning), and when it runs.
- Creating and editing a filterEvery field of the Tender Navigator filter editor explained: Basics, Search criteria, AI evaluation, Notifications & schedule, Usage, Tender budget, the setup checklist, and the three ways to start a filter.
- Limitations of AI evaluationWhat Tender Navigator's AI features can and cannot do, drawn from its AI Transparency Notice: advisory outputs, language and jurisdiction accuracy, regulatory recency, technical specifications, training bias and hallucination.