Foundational Models

Clear Ideas integrates state-of-the-art language models from leading AI providers including OpenAI, Anthropic, Google, Cohere, and xAI. These foundational models power all AI features within the platform, enabling everything from conversational chat to complex multi-step workflows.

Model Architecture Overview

Foundational models are large language models trained on extensive datasets to understand and generate human-like text. Clear Ideas provides access to multiple model families, each optimized for different use cases:

• Conversational Models: Designed for natural dialogue, quick responses, and interactive experiences
• Reasoning Models: Built for complex analysis, problem-solving, and structured thinking
• Specialized Models: Optimized for specific domains like coding, data analysis, or creative writing

The platform includes both cloud-hosted models from major providers and high-performance models hosted on Groq's optimized inference infrastructure, ensuring fast response times and cost efficiency.

Supported Models

The models table identifies image-capable models and includes per-image output pricing where applicable.

Model and Reasoning Selector

Clear Ideas uses one model selector for model choice and reasoning effort. Choose Intelligent when you want Clear Ideas to pick the best model for the task, or choose a specific model when you need predictable behavior, pricing, or capability support.

The reasoning setting controls how much deliberate reasoning budget the model should use:

• Auto uses Clear Ideas' default behavior for the selected model and provider. It is recommended for most work.
• Low favors speed and lower cost.
• Medium balances speed and depth.
• High and Extra High spend more reasoning budget for difficult analysis, planning, coding, or synthesis.

Auto often behaves like a balanced medium setting on current reasoning-capable models, but it is not a fixed medium override. It means Clear Ideas does not force a specific effort level unless the selected model or workflow requires one.

AI Chat uses concrete model choices because chat runs immediately. AI Workflows can also offer latest aliases so future scheduled runs can follow the current production model. Choose a concrete model version when repeatability matters more than automatic upgrades.

Organization model policy can allow all models, latest aliases, or specific concrete models. Deprecated models may remain visible only when they already exist in policy or older workflow definitions so administrators can remove or replace them deliberately. Deprecated models are not automatically replaced by successors at execution time. If a workflow or job references a deprecated or no-longer-permitted model, update it to an active permitted model before running it.

Model Capabilities and Selection

Reasoning vs. Conversational Models

Clear Ideas provides models optimized for different interaction patterns and cognitive workloads. Understanding these distinctions helps in selecting the most effective model for your specific use case.

Reasoning Models excel at structured thinking and complex problem-solving:

• Multi-step analysis and logical deduction
• Technical problem-solving and code generation
• Mathematical computations and data analysis
• Research synthesis and detailed explanations

These models, such as OpenAI's GPT-5.5 pro and Anthropic's Claude Opus series, allocate more computational resources to deliberate processing, resulting in higher accuracy for complex tasks but potentially longer response times.

Conversational Models prioritize natural interaction and rapid responses:

• Natural dialogue and contextual understanding
• Quick Q&A and information retrieval
• Interactive brainstorming and ideation
• Concise summaries and explanations

Models like GPT-5.4-mini and Claude Haiku are optimized for conversational flow, making them ideal for chat interfaces and time-sensitive interactions where responsiveness is critical.

Specialized and High-Performance Models

Beyond general-purpose models, Clear Ideas offers specialized options:

Code-Optimized Models: Models such as GPT-5.3-Codex specialize in programming tasks, offering strong performance for software development, debugging, and technical implementation.

Groq-Hosted Models: Select open-source models are hosted on Groq's advanced inference infrastructure, providing exceptional speed and efficiency. These include GPT-OSS variants that deliver enterprise-grade performance with optimized latency.

Multimodal Models: Google's Gemini series integrates text, vision, and multimodal understanding, enabling more comprehensive analysis of diverse content types.

Intelligent Model Selection

Clear Ideas features an Intelligent model selector that automatically optimizes model choice based on contextual analysis. The system evaluates:

• Task Complexity: Determines reasoning depth requirements and selects appropriate model capabilities
• Content Type: Adapts to text, code, data analysis, or multimodal content
• Response Parameters: Considers desired length, detail level, and output format
• Performance Constraints: Balances speed, cost, and accuracy based on user preferences
• Cost Optimization: When multiple models are viable, prioritizes the most cost-effective option

This intelligent routing ensures optimal performance without requiring manual model selection for most use cases.

Model Selection Considerations

While the Intelligent selector handles most scenarios, understanding key factors can help you make informed choices for specialized requirements:

Task Complexity

• Simple Tasks (summarization, basic Q&A): Conversational models like GPT-5.4-nano or Claude Haiku provide fast, cost-effective results
• Complex Analysis (research, technical problem-solving): Reasoning models like GPT-5.5 pro or Claude Opus deliver deeper analysis and higher accuracy

Response Characteristics

• Concise Output: Faster models optimize for brevity and quick responses
• Detailed/Exhaustive: Reasoning-focused models excel at comprehensive explanations and multi-step analysis

Performance Requirements

• Speed-Critical: Groq-hosted models and lightweight variants prioritize low latency
• Quality-Critical: Flagship models from each provider offer maximum capability for demanding applications

Content Type

• Code/Technical: Specialized models like Grok Code Fast 1 provide superior programming assistance
• Multimodal Data: Google's Gemini series handles diverse content types including images and documents
• Enterprise Scale: High-performance hosted models ensure consistent performance under load

File Generation Skills

Some models support file generation skills for spreadsheets, documents, and presentations. These skills require the model to follow specialized instructions, inspect source file structure when needed, call tools, and write Python code that produces the requested file.

File helpers only appear when the selected model supports skills. Stronger reasoning and coding-capable models are usually better for complex workbooks, polished documents, and presentation decks. Simpler models may still generate basic files, but output quality and repair reliability can vary.

See File Generation in Chat and Workflows.

Image Inputs in Workflows

For AI Workflows, there are two different image input paths:

• Inline {{imageRefVariable}} in prompt content uses OCR/extracted text context
• Prompt image attachments (selected imageRef variables) send actual image binaries to the model

Choose a vision-capable multimodal model when using prompt image attachments. Inline image references remain text-based and do not require vision input support.