AI Models and API keys

AI Models and API keys

Chatwith supports multiple AI models so you can balance answer quality, speed, capabilities, and message-credit cost. You can change the model without retraining knowledge sources.

Choose a model

  1. Open Settings → General.
  1. Open the model picker. Use Recommended, select a provider, or search by name.
  1. Review the Vision, Reasoning, and Files icons and the relative cost indicator.
  1. Select a model and save. Star frequently used models to keep them in Favorites.
Premium models are visible but locked when the current plan does not include model switching.

New model message-credit pricing

The multiplier below is the number of message credits charged for one AI reply.
Provider
Model
Credits per AI reply
Relative cost
OpenAI
GPT-5.6 Sol
15
$$$
OpenAI
GPT-5.6 Terra
10
OpenAI
GPT-5.6 Luna
2
$
Anthropic
Claude Opus 5
15
$$$+
Anthropic
Claude Sonnet 5
10
$$$
xAI
Grok 4.5
5
💡
Example: 100 GPT-5.6 Terra replies use 1,000 message credits. Reasoning effort does not change the multiplier, although deeper reasoning can take longer.

Other supported models

Provider
Model
Credits per AI reply
OpenAI
GPT-5.5
15
OpenAI
GPT-5.4
10
OpenAI
GPT-5.4 mini
2
OpenAI
GPT-5.3
5
OpenAI
GPT-5.2
10
OpenAI
GPT-5
10
OpenAI
GPT-5 mini
2
Anthropic
Claude Opus 4.8, 4.7, or 4.6
15
Anthropic
Claude Sonnet 4.6 or 4.5
10
Google
Gemini Flash 2.5
3
Mistral
Mistral Medium
2

Which new model should I use?

  • GPT-5.6 Sol: strongest OpenAI option for complex work where quality matters more than cost.
  • GPT-5.6 Terra: balanced OpenAI model for general support and sales.
  • GPT-5.6 Luna: fast, economical choice for high-volume conversations.
  • Claude Opus 5: premium Anthropic reasoning; monitor credit use closely.
  • Claude Sonnet 5: strong Anthropic model with a lower multiplier than Opus.
  • Grok 4.5: xAI reasoning at a mid-range multiplier.

Set reasoning effort

When supported, a second control appears beside the model picker. Instant turns reasoning off for the fastest response. Higher levels give the model more room to reason and usually increase latency.
  • GPT-5.6 Sol, Terra, and Luna: Instant, Low, Medium, High, X-High, Max.
  • Claude Opus 5 and Sonnet 5: Instant, Low, Medium, High, X-High, Max.
  • Grok 4.5: Low, Medium, High.
  • When switching models, Chatwith keeps the closest supported effort level.

Capabilities

The picker marks whether each model supports:
  • Vision: understands uploaded images.
  • Reasoning: can spend additional effort on complex requests.
  • Files: reads uploaded PDFs.

Use your own OpenAI API key

You can add an OpenAI API key in Settings → General for supported OpenAI models. Your OpenAI account must have access and billing enabled for the model. Keys from other providers are not accepted in this field.

Related guides