> ## Documentation Index > Fetch the complete documentation index at: https://docs.chatnorris.ai/llms.txt > Use this file to discover all available pages before exploring further. # AI models and credits > Which AI models are available in ChatNorris, how many credits each one uses, and which ones are included in your plan. # AI models and credits ChatNorris lets you choose the artificial intelligence model each agent uses to process responses. Models vary in capability, speed, and cost in credits. ## What are credits? Each plan includes a monthly pool of **AI credits**, not a fixed number of "messages". Each model has a different cost in credits per generated response — more advanced models use more credits, simpler and faster ones use fewer. You choose the model based on what you need; ChatNorris calculates and deducts the usage automatically. > [!NOTE] > > There's a second, separate pool of **resync credits**, used only when a knowledge base source updates automatically (auto-sync). It doesn't compete with the AI pool. See [Knowledge Base](https://docs.chatnorris.ai/en/guias/knowledge-base.md#automatic-sync). ### Credit pools by plan | Plan | AI credits / month | Resync credits / month | Email credits / month | | - | - | - | - | | **Free** | 300 | — (auto-sync not available) | 25 | | **Standard** | 3,000 | 15,000 | 1,000 | | **Premium** | 15,000 | 50,000 | 4,000 | ## Email credits There's a third, separate pool of **email credits**, deducted each time the agent successfully sends an email — either: * a reply sent through the [Email channel](https://docs.chatnorris.ai/en/guias/canal-email.md), or * a send from the [Send Email](https://docs.chatnorris.ai/en/guias/acciones.md#send-email) action. It costs **1 credit per successfully sent email**. A failed send attempt doesn't use a credit. Generating the reply (the AI step) still uses credits from the AI pool like any other channel — the email credit is a separate charge, only for the send. > [!NOTE] > > The Send Email action is available on all plans. The Email channel requires the Premium plan. See [Connect the Email channel](https://docs.chatnorris.ai/en/guias/canal-email.md) and [Agent actions → Send Email](https://docs.chatnorris.ai/en/guias/acciones.md#send-email). ## Credits for voice notes When a customer sends a [voice note](https://docs.chatnorris.ai/en/guias/canales-whatsapp-instagram.md#voice-notes), transcribing it uses credits from the **AI pool**, in addition to the credits for the agent's response. * It's charged by the audio's duration, in **30-second blocks**, and the result is rounded up to the nearest half credit. * The cost per minute depends on the transcription model. With today's default model it's **1.5 credits per minute**: a 20-second note uses 1 credit, a 50-second one 1.5, and a 3-minute one 4.5. * It's also charged when a person on your team is handling the conversation, because the note is transcribed so they can read it. * If the audio is longer than 3 minutes, it isn't transcribed or charged. * If the transcription provider fails, there's no charge. Audio that was processed but was silent does use its credits. In **Billing & usage → Usage** these charges appear in the **Voice note** category, separate from AI messages. ## Available models table | Model | Provider | Credits per response | Minimum plan | | - | - | - | - | | **Claude Haiku 4.5** ⭐ | Anthropic | 1 credit | Free | | **Claude Sonnet 4.6** | Anthropic | 3 credits | Free | | **Claude Sonnet 5** | Anthropic | 3 credits | Free | | **Claude Sonnet 5.5** | Anthropic | 3 credits | Free | | **Claude Opus 4.8** | Anthropic | 7 credits | Standard | | **GPT-4o mini** | OpenAI | 0.5 credits | Free | | **GPT-4o** | OpenAI | 3 credits | Free | | **GPT-5.1** | OpenAI | 3 credits | Free | | **GPT 6 Luna** | OpenAI | 1.5 credits | Standard | | **GPT 5.6 Terra** | OpenAI | 3 credits | Standard | | **GPT 6 Sol** | OpenAI | 3 credits | Standard | | **GPT 6 Astra** | OpenAI | 15 credits | Premium | | **Gemini 3.5 Flash Lite** | Google | 0.5 credits | Standard | | **Gemini 3.6 Flash** | Google | 1.5 credits | Standard | | **Gemini 3.8 Flash** | Google | 1.5 credits | Standard | | **DeepSeek V4 Flash** | DeepSeek | 0.5 credits | Standard | | **DeepSeek V4 Pro** | DeepSeek | 0.5 credits | Standard | | **Kimi K3** | Moonshot | 3 credits | Standard | | **Grok 4.5** | xAI | 3 credits | Standard | | **Grok 4.7** | xAI | 3 credits | Standard | ⭐ Recommended default model. > [!NOTE] > > The model catalog is managed from the admin panel and can change (new models, cost or minimum plan adjustments) without an app update. This table reflects the catalog in effect as of October 12, 2026. Some models aren't available in the Playground (Claude Opus 4.8, GPT 6 Sol, GPT 6 Astra, Gemini 3.8 Flash, and Grok 4.7) — if you see a model that isn't listed here, check the selector inside the agent for up-to-date details. ## Which model should you choose? ### For most cases: Claude Haiku 4.5 It's the recommended default model: fast, affordable (1 credit), and accurate for standard customer service. ### For simple queries and high volume: GPT-4o mini 0.5 credits per response — the most affordable option for straightforward FAQs and high volume. ### For more elaborate answers: Claude Sonnet or GPT When queries require a deeper understanding of context or more complete answers: * **Claude Sonnet 4.6, Sonnet 5, or Sonnet 5.5** — 3 credits, high accuracy in Spanish. They're the recommended models for actions that create something real (tickets, bookings, payments) or reveal verified data. * **GPT-4o or GPT-5.1** — 3 credits, general-purpose models from OpenAI. ### Additional models (Standard plan or higher) From the Standard plan on, you also get Claude Opus 4.8, GPT 5.6 Terra, GPT 6 Luna, and GPT 6 Sol, plus the models from Google (Gemini), xAI (Grok), Moonshot (Kimi), and DeepSeek, with different balances of cost and capability. GPT 6 Astra, the most advanced model in the catalog (15 credits), requires Premium. You can test most of them from the Playground before assigning them to an agent in production. ## How to change an agent's model You can change the model at any time without recreating the agent: **1. Go to the agent** In the side menu, click **My agents** and then the agent you want to change. **2. Find the AI settings section** You'll find the model selector next to the temperature and max tokens parameters (`My agents → [your agent] → Settings → AI`). **3. Choose the new model** Select the model you want in the dropdown. Models that require a higher plan appear locked if your plan doesn't include them. **4. Save your changes** Click **Save**. The new model applies to new conversations immediately. ## Where to see your credit usage > [card] Agent Dashboard > > `My agents → [your agent] → Dashboard` shows the AI credits for the last 7 days versus the previous 7. The general Dashboard shows the month's AI credits per agent. > [card] Analytics > > In `Analytics`, the **AI performance** tab shows the credits used per day with a projection for the month, and the **All agents** view compares each agent's credits. > [card] Billing & usage → Usage > > **Summary** shows each pool (AI, email, and resync) and the plan limits; **AI credits** breaks down usage by day, agent, model, and channel, with CSV export. See [Billing and payments](https://docs.chatnorris.ai/en/cuenta-y-facturacion/facturacion-y-pagos.md#see-the-current-month’s-usage). > [card] Per-agent limit > > `My agents → [your agent] → Settings → Limits` lets you set (optionally) a monthly credit cap for that specific agent, with its own counter of actual usage. ## Frequently asked questions > [accordion] Can I have different models in different agents? > > Yes. Each agent has its own configured model. You can have a support agent using Claude Haiku 4.5 (speed) and a sales agent using Claude Sonnet 4.6 (greater capability). > [accordion] Does changing the model affect ongoing conversations? > > No. Conversations already in progress keep the previous context, but the new model is used starting with the visitor's next message. > [accordion] Why are some models not available? > > Models that require the Standard plan or higher appear locked if your plan doesn't include them. Check your plan in **Billing & usage → Subscription**. > [accordion] I no longer see GPT-4.1 or Claude Opus 5.5 in the selector for a new agent. What happened? > > Both models are in "hidden" status: they aren't offered for new agents, but if you had an agent configured with one of them, it keeps working normally. You don't need to do anything. > [accordion] What happens if ChatNorris retires a model my agent uses? > > When a model is retired, ChatNorris schedules the migration of the agents that use it to a replacement model and notifies Owners and Admins in advance, through the bell and by email, with the date of the change. The agent also shows a notice in its settings. After the migration you can choose another model whenever you want. > [accordion] What happens if I downgrade and have an agent with a model that's no longer included? > > ChatNorris automatically reassigns that agent to a model included in your new plan (by default, an affordable one) instead of letting it stop working. You can change it manually at any time. > [accordion] Does the Playground use credits? > > No. Testing the agent from the **Playground** is free, doesn't deduct from your credit pool, and isn't recorded in Analytics. > [accordion] Do Anthropic, OpenAI, Google, xAI, Moonshot, and DeepSeek have access to my conversations? > > Messages are processed by the APIs of the provider of the model you chose, solely to generate the response. See our [Privacy Policy](https://chatnorris.ai/privacy) for more details on how this information is handled. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.