AI Models Guide

Explore how IteraOS dynamically routes your requests across 20 AI models from 4 providers with intelligent fallbacks and safety mechanisms

20

AI Models

4

Providers

7

Fallback Chains

5

Safety Layers

AI Providers & Models

🔵 OpenAI (GPT-5.x)

7 models including Sol, Terra, Luna, and legacy options. Fallback: Sol → Terra → Gemini 3.5-Flash

Strongest models: GPT-5.6-Sol (1.05M token context, $5/$30 per 1M), GPT-5.6-Terra (best value, $2.5/$15)

🔶 Anthropic (Claude)

5 models from lightweight Haiku to flagship Opus. Fable 5 is the system's advisor model.

Strongest models: Claude Fable 5 (advisor, $10/$50), Claude Opus 4.8 (reasoning, $5/$25), Claude Sonnet 5 (best value, $2/$10)

🟡 Google Gemini

4 models optimized for speed and cost. Gemini 3.5-Flash is the default Google model.

Strongest models: Gemini 3.5-Flash (fastest, $1.5/$9), Gemini 2.5-Flash-Lite (cheapest overall, $0.1/$0.4)

⚡ xAI (Grok)

7 models including reasoning, multi-agent, and code-focused variants. All accept text + image input.

Strongest models: Grok 4.6 (flagship, $2/$6), Grok Build 0.1 (code optimization, $1/$2)

🔄 Pricing Note: All prices listed are per 1 million tokens (input/output). IteraOS automatically routes requests to the most cost-effective model for each task type while maintaining quality standards. See the cost optimization section below for how you save up to 250x on AI inference.

🎯 Dynamic Model Routing

IteraOS analyzes each request in real-time and selects the optimal model based on multiple factors:

Task Complexity
Simple queries → Gemini Flash-Lite ($0.10 in) • Complex reasoning → Claude Opus ($5 in). The system matches complexity to capability.

Context Requirements
Long context windows (100K+ tokens) → GPT-5.6-Sol or Claude Opus. Short summaries → smaller models.

Cost Sensitivity
Budget-aware routing automatically downshifts to cheaper models when equivalent quality is achievable (often 10-100x savings).

Provider Availability
Real-time health checks detect rate limits, timeouts, and auth failures. If OpenAI is rate-limited, Anthropic or Google takes over automatically.

Safety Concerns
Content that triggers Claude's safety classifiers automatically routes to Opus 4.8 for human review. You stay safe and compliant.

🛡️ Safety & Governance Layers

IteraOS implements five independent layers of safety to ensure agentic automation is trustworthy and transparent:

1. Human-in-the-Loop Automation

Agentic automation requires explicit user approval before ANY system mutation. No autonomous execution without approval. You're always in control.

2. Claude Fable 5 Advisor Strategy

For high-stakes operations, Claude Fable 5 reviews the plan before execution. Fable 5 validates logic, catches errors, and suggests improvements.

3. Safety Classifier Fallback

When Claude's safety classifiers block a request, it automatically routes to Opus 4.8 for human-reviewed secondary processing. User is notified of the block with details.

4. Confidence Thresholds

Intent classification must meet 85%+ confidence before proposing actions. Below that, the request is flagged for manual review instead of auto-execution.

5. Audit Trail & Rollback

Every AI decision and system mutation is logged with full context. Users can review what happened and roll back any decision within 30 days.

💰 Cost Optimization & Model Selection

IteraOS uses intelligent routing to cut your AI inference bill by 10-100x without sacrificing quality:

📊 Real Example: Monthly Finance Report

Without Optimization: Claude Opus ($5 in) × 1M tokens = $5,000

With IteraOS Routing:

• Retrieve cached transaction summaries (no inference)
• Classify transactions (Gemini Flash-Lite: $0.10 in)
• Generate report (Claude Haiku: $1 in)
Total: $20 vs. $5,000 ✓ 250x savings

Ready to Optimize Your AI?

Join teams that are cutting AI costs by 10-100x while maintaining or improving quality. IteraOS handles model selection, fallbacks, and safety. You focus on your work.