Lewati ke konten utama

Models

The Perfect10 Pro - Enterprise Platform supports a comprehensive range of AI models from leading providers such as OpenAI, Google DeepMind, and Anthropic.
These models power diverse functionalities, from natural language understanding and reasoning to multimodal processing and external tool invocation.


Model Categories

1. Generative Large Language Models (LLMs)

Models designed for natural conversation, analytical reasoning, content generation, and code development.
These models serve as the foundation for most AI agents within the platform.

2. Retrieval-Augmented Generation (RAG)

Combines large language model capabilities with external data retrieval to provide responses grounded in enterprise-specific or domain-specific knowledge.

3. Function Calling / Tool Invocation

Models capable of executing structured tasks, such as retrieving data through APIs, triggering workflows, or integrating with external systems.

4. Multimodal Models

Models that process multiple data types—including text, images, and structured documents—to support tasks such as visual analysis, report extraction, and contextual understanding.


OpenAI Models

Recognized for advanced reasoning, precision, and versatility across creative and technical applications.

GPT-4.1

  • Provider: OpenAI
  • Description: Flagship model for advanced instruction following, complex reasoning, and large-context comprehension.
  • Ideal For: Analytical workflows, technical documentation, and complex chat orchestration.
  • Strengths: High precision, extensive context length.

GPT-4.1 Mini

  • Description: A balanced model delivering performance comparable to GPT-4o at lower latency and cost.
  • Ideal For: Cost-efficient agents requiring accurate and responsive output.

GPT-4.1 Nano

  • Description: Fastest and most lightweight model in the GPT-4.1 family.
  • Ideal For: High-frequency, low-latency applications.

GPT-4o ("Omni")

  • Description: Multimodal model capable of processing both text and images, producing fast and context-rich responses.
  • Ideal For: Conversational and visual analysis tasks.
  • Strengths: Low latency, basic visual understanding.

GPT-4o Mini

  • Description: Compact multimodal model offering over 60% cost reduction compared to GPT-3.5 Turbo.
  • Ideal For: Lightweight, interactive agents.

GPT-5 Mini

  • Description: Streamlined variant of GPT-5, optimized for lightweight reasoning and balanced accuracy.
  • Ideal For: Task automation and real-time interaction.

GPT-5 Nano

  • Description: Smallest and fastest variant, optimized for ultra-low latency environments such as developer tools.
  • Ideal For: Continuous, high-throughput agent operations.

Google DeepMind Models (Gemini)

Offers scalable, multimodal performance for analytical, scientific, and interactive applications.

Gemini 2.5 Pro

  • Provider: Google DeepMind
  • Description: Flagship reasoning and analysis model, supporting complex problem solving across domains including science and mathematics.
  • Features: Advanced multimodal comprehension (text, image, and video).
  • Ideal For: Research, data science, and analytical automation.

Gemini 2.5 Flash

  • Description: High-performance model optimized for rapid reasoning, code generation, and scientific computation.
  • Ideal For: Low-latency technical workloads.

Gemini 2.5 Flash Lite

  • Description: Lightweight version focused on cost efficiency and fast response times.
  • Ideal For: High-volume applications and scalable deployment.

Gemini 2.0 Flash

  • Description: Offers significantly reduced response latency while maintaining accuracy comparable to larger models.
  • Ideal For: Interactive and dynamic chat experiences.

Gemini 2.0 Flash Lite

  • Description: Economic variant designed for minimal latency and token efficiency.
  • Ideal For: Budget-conscious use cases.

Gemini 1.5 Pro

  • Description: Multimodal foundation model supporting text, image, and structured data inputs.
  • Ideal For: Visual and data-intensive analytical workflows.

Gemini 1.5 Flash

  • Description: Reliable multimodal model for visual understanding, summarization, and creative generation.
  • Ideal For: Visual inspection and document interpretation.

Gemini 1.5 Flash 8B

  • Description: Optimized for small prompt workloads such as chat, transcription, and translation.
  • Ideal For: Rapid, conversational use cases.

Anthropic Models (Claude)

Focused on ethical reasoning, safety, and contextually aware responses.

Claude 3 Haiku

  • Provider: Anthropic
  • Description: Compact model optimized for responsiveness and general-purpose reasoning.
  • Ideal For: Real-time interactions and customer-facing applications.

Claude 3.5 Haiku

  • Description: Enhanced version offering improved accuracy, reasoning capability, and function calling support.
  • Ideal For: Safe, conversational, and compliant enterprise environments.
  • Strengths: Fast, natural, and contextually appropriate dialogue.

Model Comparison

ModelProviderPrimary Use CaseStrengthsLimitations
GPT-4.1OpenAIReasoning, developmentHigh accuracy, long contextHigh cost, slower speed
GPT-4oOpenAIMultimodal conversationFast, image understandingLimited tool access
GPT-4o MiniOpenAILightweight multimodal tasksAffordable, responsiveReduced capability
GPT-5 NanoOpenAIUltra-low latencyFastest response timesLimited reasoning depth
Gemini 2.5 ProGoogleAnalytical reasoningMultimodal, efficientRestricted access
Gemini 2.5 FlashGoogleFast reasoning workloadsLow latency, high throughputShorter context window
Claude 3.5 HaikuAnthropicSafe conversation handlingEthical, natural dialogueLimited extensibility

Model Selection Guidelines

  • Use GPT-4.1 for advanced reasoning, complex workflows, or knowledge-intensive agents.
  • Use GPT-4o or Gemini Flash for fast, low-latency conversational tasks.
  • Use Claude 3.5 Haiku for safe, compliant, and human-like dialogue.
  • Use Gemini 2.5 Pro for data-driven analysis and large-scale scientific reasoning.
  • Use GPT-5 Nano for rapid response automation or developer tooling.

Technical Notes

  • Each agent within Perfect10 can be configured to use a specific model independently.
  • Model selection can be defined manually or programmatically (via JSON configuration).
  • Multi-model workflows are supported, enabling chained processes such as retrieval → generation → execution.
  • Multimodal input capabilities vary based on provider and version availability.