Generative AI providers and models

AIsuru supports multiple AI providers (OpenAI, Anthropic, Mistral) with language models of various sizes and capabilities. Choose the one that best fits your performance, budget, and feature requirements. This guide will help you navigate the available options and make an informed decision.

Provider overview

AIsuru integrates with several generative AI providers:

• OpenAI: use your own API key to access GPT models (including fine-tuned ones);

• Anthropic: use your own API key to access all Claude models;

• Mistral: use your own API key to access all models;

• Microsoft Azure (Azure OpenAI): use your own API key to access OpenAI's GPT models via Azure;

• Amazon Bedrock (AWS Anthropic): use your own API key to access Claude models via Amazon Bedrock;

• Google Vertex AI (Vertex Anthropic): use your own API key to access Claude models via Google Vertex AI;

• OpenAI-compatible LLMs (Custom): connect to custom providers or local servers that are compatible with the OpenAI protocol.

The choice of provider depends largely on which model you want to use.

Model overview

Currently, all AIsuru users can access the latest models of:

Provider	Model	Size
Vertex Anthropic	claude-3-7-sonnet-20250219	Large	200,000
Vertex Anthropic	Claude 4.5 Sonnet	Large	200,000 or 1M
Vertex Anthropic	Claude 4.6 Sonnet	Large	1M
Vertex Anthropic	claude-haiku-4-5-20251001	Small	200,000
Mistral	mistral-large-2407	Large	128,000
OpenAI	gpt-4o	Large	128,000
OpenAI	gpt-4o-mini	Small	128,000
OpenAI	gpt-5	Large	128,000
OpenAI	GPT-5.3 istant	Large	128,000

• OpenAI (GPT series);

• Anthropic (Claude Haiku and Sonnet);

• Mistral (Small and Large).

Reasoning models

Some models have advanced reasoning capabilities. These models show their reasoning process only in the "Conversations" tab by default.

This default behavior is designed to protect sensitive information: the reasoning process can expose internal data, processing logic, or content that shouldn't be visible to end users.

You can make the reasoning visible to end users by modifying your sharing layout.

Sometimes, providers prefer not to show the reasoning itself, so it could be hidden even if the reasoning should be showed according to your Agents layout settings.

What is context

Context represents the maximum number of tokens that can be used in each request (or question) sent to the language model. To simplify: 1 token is roughly equal to 4 characters in English. It can be personalized only for the Custom provider.

How to choose a model

The choice of language model depends on several factors. This page provides some general guidance, but you'll need to test different models with your own Agent — that's the only way to make sure it behaves correctly across all your use cases.

Here's how to pick the right model for each configuration.

Q&A configuration and expert groups

For user interactions, consider the following:

• Response complexity:

  • Complex, detailed responses: large models;

  • Simple, direct responses: small models.

• Budget:

  • Tight budget: prefer small models;

  • Flexible budget: you can go with large models.

• Response speed:

  • Immediate responses: small models;

  • Higher accuracy: large models.

• Reasoning capability: for tasks requiring complex reasoning: use models with reasoning.

Import/export configuration

For importing documents, consider:

• Document complexity:

  • Unstructured documents: you'll need to use larger models;

  • Well-structured documents: small models deliver excellent results.

• Document volume: if you need to import a large number of documents, you might use a small model to keep costs down.

Deep Thought configuration

For managing conversation memory, we always recommend using a large model with reasoning. Responses may require more time to be generated and showed to the users.