# Choosing an AI model for ListingSail

Official sources checked **October 27, 2026**. These are starting recommendations for ListingSail's bounded JSON profile extraction and form-field mapping. They are our interpretation of the vendors' documented strengths, not a ListingSail benchmark or a guarantee of account access. No requests to these models were made for this guide.

## Set your model

1. Open **AI & connections** in the desktop app and select Codex, Claude or Gemini.
2. Install a compatible official CLI yourself and configure your own authentication. See [Connect AI](https://listingsail.com/connect/).
3. Enter one model ID below in **Model**. ListingSail passes it as the CLI's `--model` argument. Enter the ID alone, without a command or additional flags.
4. Run the connection test with your account, then review an extracted profile before using the choice for a campaign. Provider calls, including tests, can consume quota or incur charges.

Leaving Model blank uses the provider CLI's runtime default. It does not promise your saved user-config default: ListingSail suppresses custom configuration for inference, and Codex runs with `--ignore-user-config`.

## Codex / OpenAI

| Choice | Model value | Suggested use in ListingSail |
| --- | --- | --- |
| Balanced starting choice | `gpt-6.1-sol` | Profile extraction and mapping that require judgment across supplied facts. |
| Fast / focused | `gpt-6-luna` | Repeated extraction and straightforward classification from clear inputs. |
| Complex | `gpt-6-astra` | Difficult mappings or conflicting source material when additional reasoning is worth the usage. |

[Official OpenAI model guidance](https://learn.chatgpt.com/docs/models) documents these exact CLI model IDs and distinguishes Sol, Luna and Astra. Model availability depends on your plan, workspace, rollout and authentication route. Older saved IDs such as `gpt-5.3-codex` are deprecated for Codex with ChatGPT sign-in; use a current available model.

[Codex authentication](https://learn.chatgpt.com/docs/auth) supports ChatGPT sign-in for eligible subscription access and an API key for usage-based access. API-key usage is billed through the customer's OpenAI Platform account separately from included ChatGPT usage. OpenAI recommends API-key authentication for programmatic CLI workflows. ListingSail does not include OpenAI usage or change your authentication.

## Claude Code / Anthropic

| Choice | Model value | Suggested use in ListingSail |
| --- | --- | --- |
| Balanced starting choice | `claude-sonnet-5-5` | Everyday profile extraction and mapping with a balance of speed and capability. |
| Fast / focused | `claude-haiku-5-5` | Clear classification and extraction tasks where latency and usage matter. |
| Complex | `claude-opus-5-5` | Ambiguous requirements or difficult source interpretation. |

[Anthropic's model overview](https://platform.claude.com/docs/en/models/overview) lists these IDs and their workload strengths. Fable is aimed at demanding reasoning and long-running work; it is not our starting recommendation for a small JSON request.

[Claude Code model configuration](https://code.claude.com/docs/en/model-config) requires version **2.1.284+** for Sonnet 5.5, **2.1.293+** for Haiku 5.5 and **2.1.280+** for Opus 5.5. Use **2.1.293 or later** for this set, with all ListingSail-required restrictions supported. Aliases `sonnet`, `haiku` and `opus` can change and can resolve differently across providers; exact IDs make the requested version clear.

[Claude Code authentication](https://code.claude.com/docs/en/authentication) supports eligible Claude subscriptions and Console/API access. Subscription limits and API billing are different routes; neither is included in the ListingSail fee. [Anthropic's product conditions](https://code.claude.com/docs/en/legal-and-compliance) require unmodified Claude Code, applicable commercial terms and each end user's own authentication and direct provider billing. ListingSail does not resell or pool Claude usage.

## Gemini CLI / Google

| Choice | Model value | Suggested use in ListingSail |
| --- | --- | --- |
| Balanced starting choice | `gemini-3.8-flash` | General extraction and mappings that need more judgment. |
| Fast / focused | `gemini-3.5-flash-lite` | Straightforward, frequent extraction and classification. |
| Complex | `gemini-4-argon` | Difficult source interpretation when the account and installed CLI support it. |

Google currently lists [Gemini 3.8 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash), [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) and [Gemini 4 Argon](https://ai.google.dev/gemini-api/docs/models/gemini-4-argon) as stable text-output models supporting structured outputs. These are vendor-listed choices, not verified ListingSail model runs.

The [Gemini CLI reference](https://geminicli.com/docs/cli/cli-reference/) accepts concrete model names with `--model`; its [model configuration](https://geminicli.com/docs/cli/generation-settings/) treats an unresolved alias as a raw model name. CLI examples and aliases can lag the API catalog. An exact ID does not establish compatibility with every installed CLI or grant access: test your account and current compatible CLI before relying on it. This release also requires the CLI's `--admin-policy` restrictions; old clients fail instead of running unrestricted.

ListingSail's Gemini adapter uses your privately supplied `GEMINI_API_KEY` in the app environment. It does not reuse Google OAuth sign-in. [Google authentication](https://geminicli.com/docs/get-started/authentication/) and [quota/billing documentation](https://geminicli.com/docs/resources/quota-and-pricing/) distinguish API-key quota and paid token usage from Google-account subscription allowances. Do not assume an AI Pro/Ultra subscription or a free API tier covers these exact models.

## Versions, costs and review

A model ID selects the hosted model; a CLI version selects the installed client. Updating one does not update the other or grant account entitlement. Prefer the balanced choice first, use a faster model when reviewed results remain accurate, and try a complex model for a specific unresolved case. These are qualitative choices, not measured cost or speed comparisons.

The desktop Model field does not expose a reasoning-effort selector. ListingSail requests structured JSON with tools disabled; it does not turn the model into a general browser agent or execute generated code. Model choice does not authorize registration, publication, payment or CAPTCHA completion. Review facts and permissions as usual.

ListingSail does not automatically retry inference or switch providers. Provider CLIs may have their own model-routing behavior; aliases and runtime defaults are not a promise that one exact model will answer every request. Account limits, provider policies and directory rules still apply. The ListingSail license, AI usage and directory fees are separate costs.

For local inference, use the separate [Ollama guide](https://listingsail.com/docs/ollama/). Continue with [ListingSail documentation](https://listingsail.com/docs/) or [AI and MCP connection setup](https://listingsail.com/connect/).