Supported AI models

Box supports a variety of AI models, categorized along two dimensions: access level and capability tier.

Access levels

Level	Description	Configuration Required
Core models	Built into Box AI and available by default for all customers	None
Customer-enabled models	Require activation by Box admins in the Admin Console or a request to Box. Some models may be subject to additional terms or pricing.	Yes

Capability tiers

Tier	Description	Best For
Standard models	High-speed, cost-efficient performance	Basic summarization, Q&A, structured data extraction from shorter or simpler documents. High-volume, low-complexity use cases
Premium models	Advanced reasoning, larger context windows, better performance on complex content	Multi-step reasoning, large taxonomies, lengthy or unstructured documents, domain-specific content

A model can be both customer-enabled and premium, or core and standard. In other words, access level and capability tiers are independent categorizations (for example, models can be either capability tier regardless of access level). The two categorizations are complementary.

Using models

How to use the supported AI models:

• Get the .
• Override the AI agent configuration used in , , , endpoints.

When using the model parameter in your API calls, use the API Name visible on each model card listed below.

For example, to get the AI agent configuration for a specific model, use the parameter and provide the azure__openai__gpt_4o_mini API name. Make sure you use two underscores after the provider name.

The list may change depending on the model availability.

Core models

Box AI is powered by the following AI models. These models are integrated with Box AI to facilitate various use cases while adhering to enterprise grade standards. Below, you’ll find information about each model, including its capabilities, intended applications, and applicable usage guidelines.

Models offered in Preview mode have not been fully performance-tested at scale. You may experience variability in output quality, availability, and accuracy.

OpenAI

openai__gpt_5_4

Multimodal model for both broad general-purpose work and most coding tasks

PremiumCompatible with Box Agent

openai__gpt_5_2

Multimodal model for coding and agentic tasks across various industries

PremiumCompatible with Box Agent

openai__gpt_5_1

Multimodal model with enterprise-grade performance and adaptive reasoning

Premium

openai__gpt_5

Multimodal model with advanced reasoning and long-context understanding

Premium

openai__gpt_5_mini

Model designed for well-defined tasks and precise prompts, suitable for lightweight tasks

Standard

azure__openai__gpt_4_1

Multimodal model, highly efficient in handling complex, multi-step tasks

PremiumISMAPFedRAMP ModerateFedRAMP HighDoD IL2

azure__openai__gpt_4_1_mini

Multimodal model designed to handle lightweight tasks and precise prompts

StandardISMAPFedRAMP ModerateFedRAMP HighDoD IL2

azure__openai__gpt_4o

Multimodal model, highly efficient in handling complex, multi-step tasks

PremiumISMAPFedRAMP ModerateFedRAMP HighDoD IL2

azure__openai__gpt_4o_mini

Multimodal model for lightweight tasks and precise prompts, best suited for focused tasks

StandardISMAPFedRAMP ModerateFedRAMP HighDoD IL2

azure__openai__text_embedding_ada_002

2nd-generation embedding model for text search, code search, and sentence similarity

EmbeddingsStandardISMAPFedRAMP ModerateFedRAMP HighDoD IL2

Google

google__gemini_2_5_pro

Multimodal model with a 1 million token context window and advanced reasoning capabilities

PremiumISMAPFedRAMP ModerateFedRAMP HighDoD IL5

google__gemini_2_5_flash

Multimodal model offering well-round capabilites, including thinking capabilities

StandardISMAPFedRAMP ModerateFedRAMP HighDoD IL5

google__gemini_2_0_flash_001

Multimodal model designed for optimal high-volume, high-frequency tasks at scale

StandardISMAPFedRAMP ModerateFedRAMP HighDoD IL5

google__gemini_2_0_flash_lite_preview

Multimodal model designed to handle lightweight tasks

StandardISMAPFedRAMP ModerateFedRAMP HighDoD IL5

Anthropic

aws__claude_4_6_opus

Multimodal model for complex tasks with a 1 million token context window

PremiumCompatible with Box Agent

aws__claude_4_6_sonnet

Multimodal model for complex tasks with a 1 million token context window

PremiumCompatible with Box Agent

aws__claude_4_5_opus

Premium model combining maximum intelligence with practical performance

PremiumISMAPFedRAMP ModerateCompatible with Box Agent

aws__claude_4_5_sonnet

Model that excels at complex agents, coding, and autonomous multi-step workflows

PremiumISMAPFedRAMP ModerateCompatible with Box Agent

aws__claude_4_5_haiku

Fast and cost-efficient model with near-frontier intelligence best for low-latency applications

StandardISMAPFedRAMP ModerateCompatible with Box Agent

aws__claude_4_opus

Model that excels at coding and complex problem-solving, powering frontier agents

PremiumISMAPFedRAMP Moderate

aws__claude_4_sonnet

Model that brings frontier performance to everyday use cases

PremiumISMAPFedRAMP Moderate

aws__claude_3_7_sonnet

Model designed to enhance language understanding and generation tasks

PremiumISMAPFedRAMP Moderate

aws__claude_3_5_sonnet

Model designed to enhance language understanding and generation tasks

PremiumISMAPFedRAMP Moderate

aws__claude_3_haiku

Model tailored for various language tasks, creative writing, and conversational AI

StandardISMAPFedRAMP Moderate

aws__claude_3_sonnet

Model that brings frontier performance to everyday use cases

PremiumISMAPFedRAMP Moderate

aws__titan_text_lite

Language model for advanced text processing and complex tasks with large context

StandardISMAPFedRAMP Moderate

Meta

ibm__llama_4_maverick

Multimodal model with a 1 million token context window and advanced reasoning capabilities

Standard

ibm__llama_4_scout

Natively multimodal, mixture-of-experts (MoE) model designed to handle lightweight tasks

Standard

ibm__llama_3_2_90b_vision_instruct

Multimodal model designed to handle lightweight tasks

Standard

Mistral AI

ibm__mistral_medium_2505

High-performance enterprise model for coding and advanced reasoning

PreviewStandard

ibm__mistral_small_3_1_24b_instruct_2503

Fast open-source multimodal model with low latency

PreviewStandard

Customer-enabled models

Certain Box AI customers may enable additional AI models upon their request or upon the models otherwise being made available through their Admin Console. Use of these models may be subject to additional terms. By selecting a customer-enabled model, the customer acknowledges that their data may be processed by additional subprocessors of their choice.

Models offered in Beta mode are provided on an as-is basis.

OpenAI

openai__gpt_o3

Multimodal model, highly efficient in handling complex, multi-step tasks

BetaPremium

Google

google__gemini_3_1_pro

Model for complex tasks with advanced reasoning and a large context window

BetaPremium

google__gemini_3_1_flash_lite

Gemini multimodal model for high-speed and cost-efficient tasks

BetaStandard

google__gemini_3_pro

Multimodal model for complex tasks with a 1 million token context window

BetaPremium

google__gemini_3_flash

Multimodal model designed for speed and efficiency across a wide range of tasks

BetaPremium

xAI

xai__grok_3_beta

Multimodal model for complex tasks with a 1 million token context window

BetaPremium

xai__grok_3_mini_beta

Multimodal model for complex tasks with a 1 million token context window

BetaPremium

Default models

The following tables list the default AI models used when you do not override the agent configuration. The model that runs depends on the agent type, the provider, and the capability tier.

Default settings shown below may not be applicable for certain customers with specific configurations or requirements.

Legacy agents

Legacy agents represent the initial generation of Box AI capabilities, designed to handle basic, single-step tasks such as simple Q&A and document summarization. The table below specifies available models for a particular use case.

Use case	Provider	Model for Standard Version	Model for Advanced or Enhanced Version
Q&A for Documents, Notes, and Hubs	OpenAI (Default)
	Claude
	Llama

Box agents

Box agents are the next generation of intelligent assistants built with advanced reasoning and orchestration, capable of planning and executing complex, multi-step workflows across your entire Box content ecosystem. The table below specifies available models for each specialized agent type.

Agent	Model for Standard Version	Model for Pro or Enhanced Version
Box Agent
Extract Agent
Security Classification Agent		N/A
Threat Analysis Agent		N/A