AI Models

DatumOS provides access to multiple AI models through Vercel AI Gateway, allowing you to choose the best model for each task based on capability, cost, and performance needs.

Available Models

Claude Haiku 4.5 (Default)

Provider: Anthropic Model ID: claude-haiku-4-5-20251001

Fast and cost-effective model that’s perfect for most everyday tasks.

Key Capabilities:

• ✅ Streaming responses
• ✅ Tool calling (search, artifacts)
• ✅ Image input
• ✅ Extended reasoning
• 200,000 token context window (~150,000 words)

Best For:

• General chat and questions
• Document search and retrieval
• Simple analysis tasks
• Quick summaries
• Most day-to-day queries

Pricing:

• Input: $0.80 per million tokens
• Output: $4.00 per million tokens

Claude Sonnet 4.5

Provider: Anthropic Model ID: claude-sonnet-4-5-20250929

Most capable model with advanced reasoning, ideal for complex tasks.

Key Capabilities:

• ✅ Streaming responses
• ✅ Tool calling (search, artifacts)
• ✅ Image input
• ✅ Extended reasoning (advanced)
• 200,000 token context window (~150,000 words)

Best For:

• Complex analysis and reasoning
• Multi-step problem solving
• Detailed technical writing
• Code generation and debugging
• Advanced image analysis
• In-depth research tasks

Pricing:

• Input: $3.00 per million tokens
• Output: $15.00 per million tokens

Grok Vision

Provider: XAI Model ID: grok-2-vision-1212

Advanced multimodal model with strong vision capabilities.

Key Capabilities:

• ✅ Streaming responses
• ✅ Tool calling (search, artifacts)
• ✅ Image input (advanced)
• ❌ Extended reasoning (not available)
• 131,072 token context window (~100,000 words)

Best For:

• Image analysis and understanding
• Drawing and diagram interpretation
• Photo documentation review
• Visual inspection tasks
• OCR and text extraction from images

Switching Models

During a Conversation

Change models mid-conversation to match your needs:

1. Click the Model dropdown in the chat header
2. Select your desired model
3. Continue the conversation with the new model

The new model will have access to the full conversation history.

Setting a Default Model

Choose your preferred default model:

1. Go to Settings > AI Models
2. Select your default model
3. Click Save

New conversations will start with this model.

Per-Message Model Selection

For fine-grained control:

1. Click the model selector before sending each message
2. Choose the appropriate model
3. Send your message

This lets you optimize cost and capability message by message.

Model Selection Guide

Decision Tree

Is the task primarily visual (images, drawings)?
├─ YES → Grok Vision
└─ NO  → Continue

Does it require complex reasoning or analysis?
├─ YES → Claude Sonnet 4.5
└─ NO  → Claude Haiku 4.5 (default)

Task-Specific Recommendations

Cost Considerations

Understanding Token Usage

Tokens are pieces of text, roughly:

• 1 token ≈ 4 characters
• 1 token ≈ 0.75 words
• 100 tokens ≈ 75 words

Example message:

Find all RFIs related to structural steel in the Civic Center project

• Input: ~15 tokens
• Typical response: ~200-500 tokens
• Total cost: under $0.01 with Haiku

Cost Comparison

Typical query (500 input tokens, 1,000 output tokens):

Sonnet costs ~3.75x more per query.

Daily usage example (100 queries/day):

Managing Costs

Best Practices:

1. Start with Haiku - Use it for initial exploration
2. Upgrade strategically - Switch to Sonnet for complex tasks
3. Monitor usage - Check Settings > Usage to track spending
4. Set limits - Configure spending alerts in Settings
5. Use shorter prompts - Be concise to reduce token usage
6. Avoid redundancy - Don’t repeat information unnecessarily

Performance Characteristics

Response Times

Time to first token (95th percentile):

Throughput

Tokens per second (streaming):

Context Windows

What is a Context Window?

The context window is the maximum amount of text (input + output) a model can process in one conversation.

Context Window Sizes:

What Fits in a Context Window?

Claude models (200K tokens):

• ~500 pages of text
• ~20 full project specifications
• ~100 typical chat messages
• Entire novels with room for analysis

Grok Vision (131K tokens):

• ~325 pages of text
• ~13 full project specifications
• ~65 typical chat messages

Managing Long Conversations

When approaching context limits:

1. Start fresh - Begin a new chat for unrelated topics
2. Be concise - Avoid repeating information unnecessarily
3. Summarize - Ask the AI to summarize long conversations

Model Capabilities Comparison

Troubleshooting

Model not available

Symptoms: Model appears grayed out or shows error

Solutions:

1. Check your subscription plan includes the model
2. Verify account is in good standing
3. Try refreshing the page
4. Check status page 

Slow responses

Symptoms: Taking longer than usual to get responses

Solutions:

1. Switch to a faster model (Haiku)
2. Check your internet connection
3. Simplify your query
4. Try during off-peak hours

Quality issues

Symptoms: Responses are not detailed enough or incorrect

Solutions:

1. Try Sonnet - Upgrade from Haiku for complex tasks
2. Be more specific - Provide more context in your query
3. Break it down - Split complex questions into steps
4. Provide examples - Show what you’re looking for

Next Steps