Model ID: gpt-5.2
Provider: OpenAI
Context Window: 400,000 tokens
Max Output: 64,000 tokens
Training Data: Up to Jul 2025

Strengths:
  - Best-in-class coding capabilities
  - Superior agentic task performance
  - Extended reasoning with configurable effort
  - Large context window (400K tokens)
  - High-quality structured output

Weaknesses:
  - Premium pricing
  - Higher latency for complex tasks
  - May be overkill for simple operations

Best Use Cases:
  • Complex code generation and debugging
  • Multi-step agentic workflows
  • Advanced reasoning tasks
  • Long-form technical content
  • System design and architecture

Model ID: gpt-5-mini
Provider: OpenAI
Context Window: 400,000 tokens
Max Output: 64,000 tokens

Strengths:
  - Excellent cost/performance ratio
  - Fast response time
  - Large context window
  - Good general capabilities
  - Suitable for most common tasks

Weaknesses:
  - Less capable at complex reasoning vs GPT-5.2
  - May require more specific prompts

Best Use Cases:
  • Data extraction
  • Classification
  • Summarization
  • Simple Q&A
  • General chat
  • Content moderation

Model ID: gpt-5-nano
Provider: OpenAI
Context Window: 400,000 tokens

Strengths:
  - Fastest OpenAI model
  - Most cost-efficient
  - Still maintains GPT-5 quality baseline
  - Large context support

Weaknesses:
  - Simplified capabilities vs full GPT-5
  - Best for well-defined tasks

Best Use Cases:
  • High-volume processing
  • Real-time applications
  • Simple classification
  • Quick lookups
  • Cost-sensitive workloads

Model ID: gpt-4.1
Provider: OpenAI
Context Window: 128,000 tokens

Strengths:
  - Smartest non-reasoning model
  - Balanced performance
  - Proven reliability
  - Good multimodal support

Weaknesses:
  - Superseded by GPT-5 series
  - Higher cost than newer efficient models

Best Use Cases:
  • Legacy application support
  • When explicit reasoning not needed
  • Balanced general-purpose tasks

Model ID: claude-opus-4-5
Provider: Anthropic
Context Window: 200,000 tokens
Max Output: 64,000 tokens
Training Data: Up to Aug 2025

Pricing:
  Input:  $5 per 1M tokens
  Output: $25 per 1M tokens

Strengths:
  - Premium model with highest intelligence
  - Extended thinking capability
  - Exceptional reasoning
  - Large output capacity (64K tokens)
  - Most recent training data (Aug 2025)

Weaknesses:
  - Higher cost
  - Moderate latency
  - May be excessive for simple tasks

Best Use Cases:
  • Complex analysis and reasoning
  • Long-form content generation
  • Research and synthesis
  • Critical decision-making
  • Advanced code review

Model ID: claude-sonnet-4-5
Provider: Anthropic
Context Window: 200,000 tokens (1M beta available)
Max Output: 64,000 tokens
Training Data: Up to Jul 2025

Pricing:
  Input:  $3 per 1M tokens
  Output: $15 per 1M tokens

Strengths:
  - Outstanding coding capabilities
  - Excellent for agentic workflows
  - Extended thinking support
  - Optional 1M context window (beta)
  - Fast performance
  - Best balance of intelligence and speed

Weaknesses:
  - Premium pricing tier
  - 1M context in beta (long context pricing applies)

Best Use Cases:
  • Complex coding tasks
  • Agentic AI workflows
  • Long document analysis (with 1M context)
  • Technical writing
  • Advanced problem solving

Model ID: claude-haiku-4-5
Provider: Anthropic
Context Window: 200,000 tokens
Max Output: 64,000 tokens
Training Data: Up to Jul 2025

Pricing:
  Input:  $1 per 1M tokens
  Output: $5 per 1M tokens

Strengths:
  - Fastest Claude model
  - Most cost-effective from Anthropic
  - Near-frontier intelligence
  - Extended thinking support
  - Large context window
  - Excellent for structured tasks

Weaknesses:
  - Less capable than Opus/Sonnet for complex reasoning
  - Better for focused tasks than open-ended ones

Best Use Cases:
  • Data extraction and structuring
  • High-volume processing
  • Real-time chat applications
  • Classification at scale
  • Cost-sensitive applications
  • Quick analysis tasks

Model ID: gemini-3-pro-preview
Provider: Google
Context Window: 1,048,576 tokens (1M)
Max Output: 65,536 tokens
Inputs: Text, Image, Video, Audio, PDF

Strengths:
  - Best-in-world multimodal understanding
  - Massive 1M token context
  - Supports all input types
  - Large output capacity
  - Advanced reasoning

Weaknesses:
  - Preview status (may change)
  - Higher latency with large inputs
  - Premium pricing expected

Best Use Cases:
  • Complex multimodal analysis
  • Video and audio understanding
  • Processing entire codebases
  • Multi-document reasoning
  • Advanced research tasks

Model ID: gemini-3-flash-preview
Provider: Google
Context Window: 1,048,576 tokens (1M)
Max Output: 65,536 tokens
Inputs: Text, Image, Video, Audio, PDF

Strengths:
  - Balanced model for speed and intelligence
  - Multimodal support
  - 1M context window
  - Fast inference
  - Frontier-level capabilities

Weaknesses:
  - Preview status
  - Slightly lower quality than Pro

Best Use Cases:
  • Large-scale multimodal processing
  • Fast document analysis
  • Real-time video/audio tasks
  • High-volume workloads

Model ID: gemini-2.5-flash
Provider: Google
Context Window: 1,048,576 tokens (1M)
Max Output: 65,536 tokens

Strengths:
  - Stable production model
  - Excellent for low-latency tasks
  - High-volume processing
  - 1M context window
  - Cost-effective

Weaknesses:
  - Generation 2.5 (superseded by Gemini 3)
  - Less capable than latest models

Best Use Cases:
  • Production applications requiring stability
  • Large-scale batch processing
  • High-throughput systems
  • Cost-optimized workloads

Model ID: gemini-2.5-pro
Provider: Google
Context Window: 1,048,576 tokens (1M)
Max Output: 65,536 tokens

Strengths:
  - State-of-the-art reasoning model
  - Complex problem solving
  - 1M context support
  - Stable for production
  - Advanced thinking capabilities

Best Use Cases:
  • Complex reasoning tasks
  • Multi-step problem solving
  • Research and analysis
  • Long-context understanding

Model ID: gemini-2.5-flash-lite
Provider: Google
Context Window: 1,048,576 tokens (1M)
Max Output: 65,536 tokens

Strengths:
  - Fastest Flash model
  - Optimized for cost-efficiency
  - Still has 1M context
  - Good for simple tasks

Best Use Cases:
  • Ultra-high-volume processing
  • Cost-critical applications
  • Simple classification
  • Quick lookups

Model ID: mistral-large-latest
Provider: Mistral AI
Context Window: 128,000 tokens
Max Output: 4,096 tokens

Strengths:
  - Strong reasoning capabilities
  - Excellent multilingual support
  - European data centers (GDPR compliant)
  - Function calling support
  - Competitive pricing

Best Use Cases:
  • European/GDPR-compliant applications
  • Multilingual tasks
  • Reasoning and analysis
  • Function calling applications

Model ID: qwen-2.5-72b-instruct
Provider: Alibaba Cloud
Context Window: 32,768 tokens
Max Output: 8,192 tokens

Strengths:
  - Excellent value for money
  - Strong Chinese/Asian language support
  - Good reasoning capabilities
  - Fast inference

Best Use Cases:
  • Chinese language tasks
  • Cost-sensitive applications
  • Asian market applications
  • Bilingual applications

Model ID: llama-3.1-405b-instruct
Provider: Meta (via various hosts)
Context Window: 128,000 tokens
Max Output: 4,096 tokens

Strengths:
  - Open weights (self-hostable)
  - Strong general capabilities
  - Customizable via fine-tuning
  - Active community
  - No vendor lock-in

Best Use Cases:
  • Self-hosted deployments
  • Fine-tuning for specific domains
  • Air-gapped environments
  • Long-term cost optimization

Model ID: llama-3.1-70b-instruct
Provider: Meta (via various hosts)
Context Window: 128,000 tokens
Max Output: 4,096 tokens

Strengths:
  - Good balance of cost and quality
  - Faster than 405B variant
  - Open weights
  - Suitable for most tasks

Best Use Cases:
  • General-purpose applications
  • Cost-effective self-hosting
  • Fine-tuning base

Extraction → Claude Haiku 4.5 or Gemini Flash-Lite
  "Extract email from: [email protected]"
  Reason: Fast, accurate, cost-effective for structured data

Classification → GPT-5 nano or Gemini 2.5 Flash-Lite
  "Is this email spam or legitimate?"
  Reason: Good accuracy, minimal cost

Summarization → GPT-5 mini or Claude Sonnet 4.5
  "Summarize this article..."
  Reason: Length-dependent (short→mini, long→Sonnet)

Complex Reasoning → Claude Opus 4.5 or GPT-5.2
  "Solve this logic puzzle..."
  Reason: Superior reasoning capabilities

Coding → GPT-5.2 or Claude Sonnet 4.5
  "Write a distributed system design"
  Reason: Best code generation and system design

Multimodal → Gemini 3 Pro or GPT-5.2
  "Analyze this video and extract insights"
  Reason: Native multimodal support

Low Complexity (0.0-0.3) → Haiku, Flash-Lite, GPT-5 nano
  "What is 2+2?"
  "Extract phone number from text"

Medium Complexity (0.3-0.6) → GPT-5 mini, Sonnet, Gemini 2.5
  "Summarize this 500-word article"
  "Categorize customer feedback"

High Complexity (0.6-1.0) → GPT-5.2, Opus 4.5, Gemini 3
  "Design a distributed architecture"
  "Analyze complex multi-modal data"

Budget: $0.001
  → Gemini Flash-Lite or GPT-5 nano

Budget: $0.005
  → GPT-5 mini, Claude Haiku, or Gemini 2.5 Flash

Budget: $0.02+
  → Any model (GPT-5.2, Opus 4.5 available if optimal)

No budget specified
  → Optimal model selected regardless of cost

{
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "text", "text": "What's in this image?"},
        {"type": "image_url", "image_url": "https://example.com/image.jpg"}
      ]
    }
  ]
}

{
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "text", "text": "Transcribe and summarize this video"},
        {"type": "video_url", "video_url": "https://example.com/video.mp4"}
      ]
    }
  ],
  "model": "gemini-3-pro-preview"
}

OpenAI: 99.8%
  Latest Models: GPT-5.2, GPT-5 mini, GPT-5 nano
  Incidents: Minimal downtime

Anthropic: 99.9%
  Latest Models: Claude Opus/Sonnet/Haiku 4.5
  Incidents: Excellent reliability

Google: 99.9%
  Latest Models: Gemini 3 Pro/Flash (preview), Gemini 2.5 series (stable)
  Incidents: Very stable

1. Primary model fails or times out
   ↓
2. Retry with exponential backoff (up to 3 attempts)
   ↓
3. Escalate to equivalent fallback model
   ↓
4. Try alternative provider with similar capabilities
   ↓
5. Return best available result

Recommended:
  - Let Perf auto-orchestrate based on task
  - Set cost budgets, not specific models
  - Use task type hints for better selection
  - Test different models for your use case

Avoid:
  - Always using the most expensive model
  - Always using the cheapest model
  - Ignoring Perf recommendations
  - Overriding without valid reason

Latency-Sensitive (chat apps):
  → GPT-5 nano, Claude Haiku 4.5, or Gemini 3 Flash

Cost-Sensitive (high volume):
  → Gemini 2.5 Flash-Lite or GPT-5 nano with auto-orchestration

Quality-Critical (customer-facing):
  → Claude Opus 4.5 or GPT-5.2 with validation

Coding & Agentic:
  → GPT-5.2 or Claude Sonnet 4.5

Multimodal:
  → Gemini 3 Pro or GPT-5.2

GDPR-Compliant:
  → Mistral Large (European data centers)

Balanced:
  → Perf auto-orchestration (learns optimal mix)

GPT-5 series (400K tokens)
  Optimal: < 100K tokens
  Good: 100K-200K tokens
  Extended: 200K-400K tokens

Claude 4.5 series (200K / 1M beta)
  Optimal: < 100K tokens
  Good: 100K-200K tokens
  Extended (1M beta): 200K-1M tokens (long context pricing)

Gemini series (1M tokens)
  Optimal: < 500K tokens
  Good: 500K-800K tokens
  Extended: 800K-1M tokens

Assumptions:
  - 10,000 conversations/month
  - Average 5 messages per conversation (50K total)
  - 100 input tokens, 50 output tokens per message

Model: GPT-5 mini
  Cost per message: ~$0.00006
  Monthly cost: ~$3,000

Model: Claude Haiku 4.5
  Cost per message: ~$0.00015
  Monthly cost: ~$7,500

Model: Gemini 2.5 Flash-Lite
  Cost per message: ~$0.00001
  Monthly cost: ~$500

Model: Perf Auto-Orchestration
  Mix: 60% Flash-Lite, 25% GPT-5 nano, 10% Haiku, 5% Sonnet
  Monthly cost: ~$1,200
  Savings: $1,800/month (60%) vs GPT-5 mini baseline