Skip to main content

AI Models

Approved AI models for custom developed GenAI applications in OneAI platform.

General Requirements & Compliance
  • All models must comply with company data governance policies and security standards
  • Usage must adhere to vendor-specific terms of service and acceptable use policies
  • PII and sensitive data handling requires additional approval and encryption
  • Production deployments require security review and compliance sign-off
  • Cost monitoring and budget allocation must be configured before use
  • Models with specific legal requirements (shown in red below) require additional documentation
Vendor / Model Description Specifications When to Use Pros Cons
Google
Gemini 2.5 Pro
(v2.5)
via Google Cloud

Approved

 Google’s most advanced model with massive context window and multimodal capabilities. Excellent for complex reasoning, long document analysis, and multimodal tasks.
Context: 1M tokens
Output: 8,192 tokens
Cutoff: Jan 2025
Speed: Moderate (3–6s)
  • Long context analysis
  • Multimodal tasks
  • Complex reasoning
  • Research & analysis
  • Massive 1M context
  • Excellent multimodal
  • Latest knowledge
  • Strong reasoning
  • Advanced capabilities
  • Higher cost
  • Moderate speed
  • Newer model
  • Less ecosystem maturity
Google
Gemini 2.5 Flash
(v2.5)
via Google Cloud

Approved

 Fast and efficient Gemini model with large context window. Optimized for speed while maintaining strong capabilities across text and multimodal tasks.
Context: 1M tokens
Output: 8,192 tokens
Cutoff: Jan 2025
Speed: Fast (1–3s)
  • Fast processing
  • Long documents
  • Cost optimization
  • Real-time apps
  • High-volume tasks
  • Very fast
  • 1M context window
  • Cost-effective
  • Good multimodal
  • Balanced performance
  • Less capable than Pro
  • Newer model
  • Limited complex reasoning
Amazon Bedrock
Anthropic Claude
(3.5 Sonnet)
via AWS Bedrock

Approved

 Flagship model balancing intelligence, speed, and cost. Strong at analysis, coding, content creation, and reasoning with nuanced understanding.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: Apr 2024
Speed: Fast (2–4s)
  • Complex reasoning
  • Code review
  • Content creation
  • Data analysis
  • Long conversations
  • Safety-critical apps
  • Excellent reasoning
  • Strong safety
  • 200K context window
  • Nuanced understanding
  • Good code generation
  • Balanced cost
  • More expensive than smaller models
  • May be cautious at times
  • Limited real-time info
  • Smaller ecosystem
  • No image generation
Amazon Bedrock
Anthropic Claude
(3.5 Haiku)
via AWS Bedrock

Approved

 Fast and efficient Claude model optimized for speed and cost. Best for high-volume processing and quick response times while maintaining quality.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: Jul 2024
Speed: Very Fast (<1s)
  • Simple tasks
  • Quick responses
  • High throughput
  • Cost-sensitive apps
  • Real-time interactions
  • Very fast
  • Low cost
  • Good for simple tasks
  • 200K context window
  • Reliable performance
  • Limited complex reasoning
  • Less capable than Sonnet
  • Simpler outputs
Amazon Bedrock
Anthropic Claude
(3.7 Sonnet)
via AWS Bedrock

Approved

 Balanced Claude Sonnet model for general-purpose tasks. Enhanced version with improved performance over 3.5.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: Oct 2024
Speed: Fast (2–4s)
  • General purpose tasks
  • Content creation
  • Code review
  • Analysis work
  • Document processing
  • Good balance
  • 200K context
  • Cost-effective
  • Improved over 3.5
  • Reliable
  • Not the latest version
  • Older knowledge cutoff
  • Less advanced than 4.x series
Amazon Bedrock
Anthropic Claude
(Sonnet 4)
via AWS Bedrock

Approved

 High-performance Claude Sonnet model with strong reasoning capabilities. Next-generation model with significant improvements.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: Mar 2025
Speed: Fast (2–4s)
  • Complex reasoning
  • Code generation
  • Analysis tasks
  • Advanced workflows
  • Enterprise applications
  • Strong performance
  • 200K context window
  • Latest knowledge cutoff
  • Advanced reasoning
  • Better code quality
  • Higher cost
  • May be overkill for simple tasks
  • Premium pricing
Amazon Bedrock
Anthropic Claude
(Sonnet 4.5)
via AWS Bedrock

Approved

 Latest iteration of Claude Sonnet with enhanced reasoning, improved coding capabilities, and better instruction following. Offers superior performance while maintaining the balanced approach of the Sonnet series.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: Jul 2024
Speed: Fast (2–4s)
  • Advanced reasoning tasks
  • Complex code generation
  • Agentic workflows
  • Multi-step problem solving
  • Technical documentation
  • Data transformation
  • Enhanced reasoning over 4.0
  • Better code quality
  • Improved instruction following
  • Strong agentic capabilities
  • 200K context window
  • Maintains safety standards
  • Best-in-class for complex tasks
  • Higher cost than earlier versions
  • May be overkill for simple tasks
  • Similar speed to 4.0
  • Limited ecosystem vs OpenAI
  • No image generation
!
Limitation #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Amazon Bedrock
Anthropic Claude
(Haiku 4.5)
via AWS Bedrock

Approved

 Fast and efficient Claude model optimized for speed and cost. Latest Haiku version with improved capabilities.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: Feb 2025
Speed: Very Fast (<1s)
  • High-volume processing
  • Quick responses
  • Cost-sensitive apps
  • Real-time interactions
  • Simple automated tasks
  • Extremely fast
  • Cost-effective
  • 200K context window
  • Latest version
  • Good for high throughput
  • Less capable than Sonnet
  • Limited complex reasoning
  • Simpler outputs
Amazon Bedrock
Cohere Command
Command R/R+
via AWS Bedrock

Approved

 Enterprise-focused models optimized for retrieval and business applications. Strong RAG capabilities and multilingual support.
Context: 128K tokens
Output: 4,096 tokens
Cutoff: Early 2024
Speed: Fast (2–4s)
  • RAG applications
  • Business tasks
  • Information retrieval
  • Multilingual processing
  • Enterprise search
  • Excellent for RAG
  • Enterprise features
  • Strong retrieval
  • Multilingual support
  • Cost-effective
  • Limited compared to GPT-4
  • Niche use cases
  • Smaller context window
Amazon Bedrock
Cohere Embed
(3)
via AWS Bedrock

Approved

 Semantic search and embedding model for document processing and similarity matching. High-quality vector representations.
Context: N/A
Output: Embeddings
Cutoff: Not applicable
Speed: Very Fast (<1s)
  • Search applications
  • Similarity matching
  • Document clustering
  • Recommendation systems
  • Semantic search
  • High-quality embeddings
  • Fast processing
  • Multilingual support
  • Cost-effective
  • Specialized purpose
  • Single purpose only
  • Not a generative model
  • Requires integration
Amazon Bedrock
Cohere Embed
(4)
via AWS Bedrock

Approved

 Enhanced embedding model with improved accuracy and better multilingual capabilities. Latest version with optimized performance.
Context: N/A
Output: Embeddings
Cutoff: Not applicable
Speed: Very Fast (<1s)
  • Advanced search
  • Cross-lingual tasks
  • Large-scale retrieval
  • Semantic analysis
  • Production embeddings
  • Enhanced embeddings
  • Improved accuracy
  • Better multilingual
  • Optimized performance
  • Latest version
  • Single purpose only
  • Higher cost than v3
  • Not a generative model
Amazon Bedrock
Cohere Rerank
(3.5)
via AWS Bedrock

Approved

 Specialized model for reranking search results and relevance scoring. Optimized for improving search quality.
Context: N/A
Output: Rankings
Cutoff: Not applicable
Speed: Very Fast (<1s)
  • Search reranking
  • Relevance scoring
  • Information retrieval
  • Search optimization
  • Result refinement
  • Excellent reranking
  • Very fast
  • Specialized purpose
  • Improves search quality
  • Easy integration
  • Single purpose only
  • Not a general LLM
  • Requires search system
Amazon Bedrock
Llama
(2)
via AWS Bedrock

Approved

 Open source model with versatile capabilities. Community-driven with good performance for general tasks.
Context: 4K tokens
Output: 2,048 tokens
Cutoff: Sep 2022
Speed: Fast (2–4s)
  • General tasks
  • Research
  • Development
  • Experimentation
  • Open source
  • Versatile
  • Community support
  • Self-hostable
  • Older knowledge
  • Limited context
  • Outdated information
  • Less capable than newer models
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Amazon Bedrock
Llama
(3)
via AWS Bedrock

Approved

 Improved version of Llama with better reasoning and enhanced performance. Stronger capabilities than Llama 2.
Context: 8K tokens
Output: 4,096 tokens
Cutoff: Mar 2023
Speed: Fast (2–4s)
  • Advanced tasks
  • Research projects
  • Development
  • Testing
  • Improved capabilities
  • Better reasoning
  • Enhanced performance
  • Open source
  • Still older knowledge
  • Smaller context than competitors
  • Limited vs GPT-4
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Amazon Bedrock
Mistral
(7B)
via AWS Bedrock

Approved

 Efficient open-source model with 7 billion parameters. Good balance of performance and resource usage.
Context: 32K tokens
Output: 4,096 tokens
Cutoff: Early 2023
Speed: Very Fast (<2s)
  • Cost-sensitive apps
  • Self-hosting
  • Simple tasks
  • High-volume processing
  • Open source
  • Very cost-effective
  • Fast processing
  • Efficient
  • Good for simple tasks
  • Limited capabilities
  • Older knowledge
  • Smaller context than newer models
  • Less sophisticated
Amazon Bedrock
Mistral
(8x7B)
via AWS Bedrock

Approved

 Mixture of Experts model with efficient 8x7B parameter architecture. Better performance than standard 7B.
Context: 32K tokens
Output: 4,096 tokens
Cutoff: Early 2024
Speed: Very Fast (<2s)
  • Efficient processing
  • Balanced performance
  • Cost optimization
  • Production workloads
  • Efficient MoE architecture
  • Good performance/cost
  • Fast processing
  • Better than 7B
  • Open source
  • Limited vs larger models
  • Smaller context
  • Less capable than premium models
GCP
Mistral OCR
(25.05)
via GCP

Pending

 Specialized optical character recognition model for document processing and text extraction.
Context: N/A
Output: Text extraction
Cutoff: Not applicable
Speed: Fast (1–3s)
  • Document scanning
  • Text extraction
  • Image to text
  • Form processing
  • Specialized OCR
  • Good accuracy
  • Fast processing
  • Multi-format support
  • Single purpose
  • Not a general LLM
  • Limited to OCR tasks
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Amazon Bedrock
OpenAI GPT OSS
(20B)
via AWS Bedrock

Approved

 Mid-size open source model with 20 billion parameters. Good balance for self-hosting scenarios.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: Jun 2024
Speed: Fast (2–4s)
  • Medium-scale tasks
  • Self-hosting
  • Balanced capability
  • Private deployment
  • Open source
  • Good balance
  • Self-hostable
  • Privacy control
  • Cost-effective
  • Requires infrastructure
  • Limited vs GPT-4
  • Smaller context
Amazon Bedrock
OpenAI GPT OSS
(120B)
via AWS Bedrock

Approved

 Large open source model with 120 billion parameters. High capability for complex tasks.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: Jun 2024
Speed: Slower (5–10s)
  • Complex tasks
  • Self-hosting
  • High capability needs
  • Advanced processing
  • Open source
  • Large capacity
  • Strong performance
  • Self-hostable
  • Privacy control
  • High infrastructure needs
  • Slower processing
  • Resource intensive
  • Complex deployment
Amazon Bedrock
DeepSeek
(R1)
via AWS Bedrock

Approved

 Advanced reasoning model optimized for complex analytical tasks, mathematical problems, and logical reasoning.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: Mid 2024
Speed: Moderate (3–5s)
  • Complex reasoning
  • Mathematical problems
  • Analytical tasks
  • Code generation
  • Technical analysis
  • Strong reasoning
  • Good analytical performance
  • Cost-effective
  • Math capabilities
  • Code generation
  • Limited context window
  • Smaller ecosystem
  • Less versatile
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
GCP
Imagen
(4)
via AWS Bedrock

Approved

 Advanced image generation model with high-quality output and customizable options. Fast rendering capabilities.
Context: N/A
Output: Images
Cutoff: Not applicable
Speed: Fast (3–5s)
  • Image generation
  • Creative visuals
  • Fast rendering
  • Marketing materials
  • Design prototypes
  • High quality images
  • Fast generation
  • Customizable
  • Good control
  • Versatile styles
  • Image only
  • Not a text model
  • Content policy restrictions
  • Usage limits
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Black Forest Labs
Flux
(.1 schnell)
via GCP

Requested

 Advanced image generation model with high fidelity and text-to-image capabilities. Professional-grade output quality.
Context: N/A
Output: Images
Cutoff: Not applicable
Speed: Moderate (4–7s)
  • Professional graphics
  • Marketing
  • Design
  • Content creation
  • High-quality visuals
  • High fidelity output
  • Text-to-image
  • Style control
  • Professional quality
  • Google integration
  • Image only
  • Higher cost
  • Moderate speed
  • Requires GCP
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Krea
Flux
(.1 krea)
via Direct API

Requested

 Fast image generation model optimized for speed. Quick iterations and rapid prototyping.
Context: N/A
Output: Images
Cutoff: Not applicable
Speed: Fast (3–5s)
  • Rapid prototyping
  • Quick visuals
  • Iterative design
  • Time-sensitive projects
  • Fast rendering
  • Fast generation
  • High quality images
  • Good control
  • Quick iterations
  • Efficient
  • Image only
  • Not a text model
  • Limited availability
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
GCP
Veo
(3, 3 Fast)
via GCP

Requested

 Krea’s variant of Flux optimized for creative image generation with artistic style control.
Context: N/A
Output: Images
Cutoff: Not applicable
Speed: Moderate (4–7s)
  • Creative projects
  • Artistic content
  • Design exploration
  • Visual research
  • Style experiments
  • Creative control
  • High quality
  • Artistic style
  • Flexible output
  • Unique aesthetics
  • Image only
  • Slower than Schnell
  • Niche provider
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Azure
GPT
(5)
via GCP

Requested

 Google’s advanced video generation model with multiple format support and quality output options.
Context: N/A
Output: Video
Cutoff: Not applicable
Speed: Slow (30–60s)
  • Video generation
  • Animation
  • Creative video content
  • Marketing videos
  • Presentations
  • Video generation
  • High quality output
  • Multi-format
  • Flexible duration
  • Advanced AI
  • Video only
  • Slow processing
  • High cost
  • Resource intensive
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
Qwen3-coder
(480B-A35B-Instruct)
via Self-hosted / Ollama

Approved

 Code-focused large model with extensive context. Self-hosted for privacy-preserving code generation.
Context: 128K tokens
Output: 8,192 tokens
Cutoff: Mid 2024
Speed: Moderate (4–8s)
  • Code generation
  • Development
  • Private projects
  • Internal tools
  • Secure environments
  • Code-focused
  • Large context
  • Self-hosted
  • Privacy-preserving
  • No data sharing
  • Requires infrastructure
  • Resource intensive
  • Complex deployment
  • Maintenance overhead
Self-hosted
Qwen3-coder
(30-A3B-Instruct)
via Self-hosted / Ollama

Approved

 Efficient coding model with fast inference. Self-hosted for internal development use.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: Mid 2024
Speed: Fast (2–4s)
  • Code assistance
  • Development tasks
  • Private deployment
  • Internal use
  • Quick coding help
  • Efficient
  • Fast inference
  • Self-hosted
  • Resource-efficient
  • Privacy control
  • Requires infrastructure
  • Smaller than 480B
  • Less capable for complex tasks
Self-hosted
Codellama
(7b, 13b, 34b, 70b)
via Self-hosted / Ollama

Pending determination

 Code generation models in multiple sizes. Open source with flexible deployment options.
Context: 16K tokens
Output: 4,096 tokens
Cutoff: Early 2023
Speed: Fast (2–5s)
  • Development
  • Code completion
  • Learning
  • Prototyping
  • Experimentation
  • Multiple sizes
  • Open source
  • Flexible deployment
  • Good code generation
  • Community support
  • Older model
  • Limited context
  • Usage unclear
  • Pending determination
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
Codegemma
(2b, 7b)
via Self-hosted / Ollama

Approved

 Lightweight coding models for fast performance. Easy deployment for code assistance.
Context: 8K tokens
Output: 2,048 tokens
Cutoff: Early 2024
Speed: Very Fast (<2s)
  • Code assistance
  • Development
  • Learning
  • Quick tasks
  • Fast completion
  • Lightweight
  • Fast performance
  • Open source
  • Easy deployment
  • Resource efficient
  • Small context
  • Limited capabilities
  • Basic features
  • Less sophisticated
Self-hosted
Codestral
(22b)
via Self-hosted / Ollama

Denied

 Advanced coding model with high quality output. Requires commercial license for use.
Context: 32K tokens
Output: 4,096 tokens
Cutoff: Mid 2024
Speed: Moderate (3–6s)
  • Professional development
  • Enterprise projects
  • Advanced coding
  • Production use
  • Advanced coding
  • High quality
  • Specialized
  • Professional features
  • Requires commercial license
  • Currently denied
  • Cost barrier
  • License complexity
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
DeepSeek-coder
(v2)
via Self-hosted / Ollama

Approved

 Strong code generation model with reasoning capabilities. Self-hosted for secure environments.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: Mid 2024
Speed: Moderate (3–5s)
  • Code development
  • Problem-solving
  • Internal projects
  • Secure environments
  • Private deployment
  • Strong code generation
  • Reasoning
  • Self-hosted
  • Privacy-focused
  • Good performance
  • Requires infrastructure
  • Moderate speed
  • Self-management needed
Self-hosted
Granite-code
via Self-hosted / Ollama

Requested

 IBM enterprise-focused code generation model. Designed for regulated environments and corporate projects.
Context: 32K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • Enterprise development
  • Corporate projects
  • Regulated environments
  • Internal tools
  • Enterprise-focused
  • Code generation
  • Self-hostable
  • IBM support
  • Not yet approved
  • Limited information
  • Pending review
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
Llama4Scout
via Self-hosted / Ollama

Requested

 Exploration variant of Llama 4. Versatile open source model for research and experimentation.
Context: 32K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • Research
  • Experimentation
  • Development
  • Testing
  • Exploration model
  • Versatile
  • Open source
  • Self-hosted
  • Not yet approved
  • Unknown capabilities
  • Limited documentation
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
Llama4Maverick
via Self-hosted / Ollama

Requested

 Advanced variant of Llama 4 with enhanced features. Open source with flexible deployment.
Context: 32K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • Advanced projects
  • Research
  • Development
  • Innovation
  • Advanced variant
  • Enhanced features
  • Open source
  • Flexible
  • Not yet approved
  • Unknown capabilities
  • Pending release
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
Llama3-gradient
via Self-hosted / Ollama

Requested

 Specialized Llama 3 variant optimized with gradient techniques. Research-focused model.
Context: 16K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • Research applications
  • Specialized tasks
  • Development
  • Experimentation
  • Specialized variant
  • Gradient-optimized
  • Self-hosted
  • Research-focused
  • Not yet approved
  • Niche use case
  • Limited info
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
Kimi-K2
via Self-hosted / Ollama

Requested

 Long context multilingual model. Self-hosted for document analysis and multilingual tasks.
Context: 200K tokens
Output: 8,192 tokens
Cutoff: TBD
Speed: TBD
  • Document analysis
  • Multilingual tasks
  • Research
  • Long-form content
  • Long context
  • Multilingual
  • Self-hosted
  • Research model
  • Not yet approved
  • Limited availability
  • Unknown performance
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
GPT-OSS
(20B)
via Self-hosted / Ollama

Requested

 Open source GPT model for self-hosting. Community-driven with flexible deployment.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • General tasks
  • Research
  • Development
  • Private deployment
  • Open source
  • Self-hosted
  • Community-driven
  • Flexible
  • Not yet approved
  • Requires infrastructure
  • Unknown capabilities
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
GPT-OSS
(120B)
via Self-hosted / Ollama

Requested

 Large open source GPT model with advanced capabilities. Self-hosted for high-performance tasks.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • Complex tasks
  • Research
  • Enterprise projects
  • Advanced applications
  • Large model
  • Advanced capabilities
  • Self-hosted
  • High performance
  • Not yet approved
  • High resource needs
  • Complex deployment
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Azure
GPT-OSS-Safeguard
(120B)
via Azure

Requested

 Safety-enhanced large GPT model with moderation capabilities. Enterprise-ready for regulated environments.
Context: 128K tokens
Output: 8,192 tokens
Cutoff: TBD
Speed: TBD
  • Regulated environments
  • Safety-critical apps
  • Compliance-focused
  • Enterprise use
  • Safety-enhanced
  • Moderation
  • Secure
  • Enterprise-ready
  • Compliance features
  • Not yet approved
  • Unknown cost
  • Pending availability
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Azure
GPT-OSS-Safeguard
(20B)
via Azure

Requested

 Efficient safeguard model with moderation. Cost-effective with security features.
Context: 64K tokens
Output: 4,096 tokens
Cutoff: TBD
Speed: TBD
  • Moderated applications
  • Safety requirements
  • Compliance needs
  • Controlled environments
  • Efficient
  • Moderation
  • Cost-effective
  • Secure
  • Not yet approved
  • Smaller than 120B
  • Limited info
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Self-hosted
BGE-BASE-EN
(1.5)
via Self-hosted

Approved

 English-focused embedding model. Self-hosted for semantic search and document retrieval.
Context: N/A
Output: Embeddings
Cutoff: Not applicable
Speed: Very Fast (<1s)
  • Semantic search
  • Document retrieval
  • Similarity matching
  • Internal applications
  • English-focused
  • Self-hosted
  • Efficient
  • Fast processing
  • Privacy control
  • Single language
  • Single purpose
  • Not generative
  • Requires infrastructure
Self-hosted
clip-ViT-B-32
(32)
via Self-hosted

Denied

 Vision-language model for image-text matching and cross-modal retrieval.
Context: N/A
Output: Embeddings
Cutoff: Not applicable
Speed: Fast (1-3s)
  • Image search
  • Cross-modal retrieval
  • Visual applications
  • Content matching
  • Vision-language
  • Image-text matching
  • Multimodal
  • Versatile
  • License not specified
  • Currently denied
  • Legal uncertainty
  • Deployment blocked
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
Amazon Bedrock
Anthropic Claude
(Opus 4.5)
via AWS Bedrock

Approved

 Premium performance Claude model with maximum capability. Advanced reasoning and enterprise features for mission-critical tasks.
Context: 200K tokens
Output: 16,384 tokens
Cutoff: Aug 2024
Speed: Moderate (4-8s)
  • Critical applications
  • Strategic projects
  • High-value tasks
  • Premium use cases
  • Complex analysis
  • Premium performance
  • Maximum capability
  • Advanced reasoning
  • Enterprise features
  • Largest output
  • Best quality
  • Highest cost
  • Slower than other models
  • May be overkill
  • Premium pricing
!
LIMITATION #1
Must include attribution in commercial applications. Data retention policy applies. Enhanced compliance monitoring required.
GCP
Gemini
(3)
via GCP

Approved

 Next-generation Gemini model with advanced multimodal capabilities and enhanced reasoning.
Context: 1M tokens
Output: 8,192 tokens
Cutoff: Feb 2025
Speed: Moderate (3-6s)
  • Advanced applications
  • Multimodal projects
  • Research
  • Innovation
  • Long context tasks
  • Next-gen model
  • Advanced multimodal
  • Enhanced reasoning
  • Improved performance
  • 1M context
  • Higher cost
  • Moderate speed
  • Very new model
  • Limited track record
GCP
NanoBanana
(1)
via GCP

Requested

 Lightweight efficient model for edge deployment and resource-limited environments.
Context: 4K tokens
Output: 1,024 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Edge deployment
  • Resource-limited
  • Quick tasks
  • Efficient processing
  • Lightweight
  • Efficient
  • Fast
  • Resource-friendly
  • Low cost
  • Limited capabilities
  • Small context
  • Basic features
  • Less sophisticated
Self-hosted
HunyuanOCR
—
via Self-hosted

Requested

 Enhanced version of NanoBanana with better performance while maintaining efficiency.
Context: 8K tokens
Output: 2,048 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Production workloads
  • Scalable solutions
  • Efficient operations
  • Balanced tasks
  • Enhanced version
  • Better performance
  • Efficient
  • Balanced
  • Cost-effective
  • Still limited vs full models
  • Smaller context
  • Basic capabilities
Self-hosted
SigLIP-base
(16-384)
via Self-hosted

Approved

 Lightweight efficient model for edge deployment and resource-limited environments.
Context: 4K tokens
Output: 1,024 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Edge deployment
  • Resource-limited
  • Quick tasks
  • Efficient processing
  • Lightweight
  • Efficient
  • Fast
  • Resource-friendly
  • Low cost
  • Limited capabilities
  • Small context
  • Basic features
  • Less sophisticated
Self-hosted
all-MinniLM-L6-v2
(2)
via Self-hosted

Approved

 Enhanced version of NanoBanana with better performance while maintaining efficiency.
Context: 8K tokens
Output: 2,048 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Production workloads
  • Scalable solutions
  • Efficient operations
  • Balanced tasks
  • Enhanced version
  • Better performance
  • Efficient
  • Balanced
  • Cost-effective
  • Still limited vs full models
  • Smaller context
  • Basic capabilities
Self-hosted
Openai/whisper-base
(20250625)
via Self-hosted

Approved

 Lightweight efficient model for edge deployment and resource-limited environments.
Context: 4K tokens
Output: 1,024 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Edge deployment
  • Resource-limited
  • Quick tasks
  • Efficient processing
  • Lightweight
  • Efficient
  • Fast
  • Resource-friendly
  • Low cost
  • Limited capabilities
  • Small context
  • Basic features
  • Less sophisticated
Self-hosted
Salesforce/blip-image
(16-Dec-25)
via Self-hosted

Approved

 Enhanced version of NanoBanana with better performance while maintaining efficiency.
Context: 8K tokens
Output: 2,048 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Production workloads
  • Scalable solutions
  • Efficient operations
  • Balanced tasks
  • Enhanced version
  • Better performance
  • Efficient
  • Balanced
  • Cost-effective
  • Still limited vs full models
  • Smaller context
  • Basic capabilities
Amazon Bedrock
Nova 2 Lite
via AWS Bedrock

Approved

 Lightweight efficient model for edge deployment and resource-limited environments.
Context: 4K tokens
Output: 1,024 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Edge deployment
  • Resource-limited
  • Quick tasks
  • Efficient processing
  • Lightweight
  • Efficient
  • Fast
  • Resource-friendly
  • Low cost
  • Limited capabilities
  • Small context
  • Basic features
  • Less sophisticated
Amazon Bedrock
Nova 2 Pro
via AWS Bedrock

Approved

 Enhanced version of NanoBanana with better performance while maintaining efficiency.
Context: 8K tokens
Output: 2,048 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Production workloads
  • Scalable solutions
  • Efficient operations
  • Balanced tasks
  • Enhanced version
  • Better performance
  • Efficient
  • Balanced
  • Cost-effective
  • Still limited vs full models
  • Smaller context
  • Basic capabilities
Amazon Bedrock
Nova 2 Omni
via AWS Bedrock

Approved

 Lightweight efficient model for edge deployment and resource-limited environments.
Context: 4K tokens
Output: 1,024 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Edge deployment
  • Resource-limited
  • Quick tasks
  • Efficient processing
  • Lightweight
  • Efficient
  • Fast
  • Resource-friendly
  • Low cost
  • Limited capabilities
  • Small context
  • Basic features
  • Less sophisticated
Amazon Bedrock
Nova 2 Sonic
via AWS Bedrock

Approved

 Enhanced version of NanoBanana with better performance while maintaining efficiency.
Context: 8K tokens
Output: 2,048 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Production workloads
  • Scalable solutions
  • Efficient operations
  • Balanced tasks
  • Enhanced version
  • Better performance
  • Efficient
  • Balanced
  • Cost-effective
  • Still limited vs full models
  • Smaller context
  • Basic capabilities
Amazon Bedrock
Nova Multimodal Embeddings
via AWS Bedrock

Approved

 Lightweight efficient model for edge deployment and resource-limited environments.
Context: 4K tokens
Output: 1,024 tokens
Cutoff: TBD
Speed: Very Fast (<1s)
  • Edge deployment
  • Resource-limited
  • Quick tasks
  • Efficient processing
  • Lightweight
  • Efficient
  • Fast
  • Resource-friendly
  • Low cost
  • Limited capabilities
  • Small context
  • Basic features
  • Less sophisticated