Skip to Content

Guide

Models

Onsomble offers 35+ AI models. Learn which one to choose for different tasks.

Different AI models have different strengths. Some are faster and cheaper. Others are more capable but cost more. This guide helps you choose.

Choosing a Model

Quick Recommendations

Use CaseRecommended ModelWhy
Daily tasksClaude Sonnet 4Great balance of speed, quality, cost
Complex analysisClaude Opus 4.5Most capable reasoning
Quick questionsClaude Haiku 4.5Fast and cheap
Code tasksGPT-4o or Claude SonnetStrong at coding
Long documentsGemini 2.5 ProLarge context window

Factors to Consider

Quality — How good are the responses?

  • Higher-tier models give better reasoning and nuance
  • Lower-tier models work fine for simple tasks

Speed — How fast do you get responses?

  • Smaller models respond faster
  • Larger models take longer but may be worth the wait

Cost — How many credits per message?

  • Ranges from 0.5 to 10 credits per message
  • Complex queries use more credits regardless of model

Context Window — How much can the model “remember”?

  • Larger windows handle longer conversations
  • Important for documents with lots of content

Available Models

Anthropic (Claude)

ModelCreditsBest For
Claude Opus 4.510Complex reasoning, difficult problems
Claude Sonnet 4.53Daily driver, great all-around
Claude Sonnet 42Balanced quality and cost
Claude Haiku 4.50.5Quick tasks, simple questions

Anthropic strengths: Nuanced reasoning, following instructions, safety

OpenAI (GPT)

ModelCreditsBest For
GPT-55Cutting-edge capability
GPT-5 Mini2Good balance
GPT-4o2Reliable all-around
GPT-4.1 Mini1Budget-friendly
o3 Mini1.5Reasoning tasks
o4 Mini2Advanced reasoning

OpenAI strengths: Coding, broad knowledge, consistency

Google (Gemini)

ModelCreditsBest For
Gemini 2.5 Pro3Long documents, analysis
Gemini 2.5 Flash1Fast responses
Gemini 3 Pro Preview4Latest capabilities

Google strengths: Large context windows, multimodal understanding

Meta (Llama)

ModelCreditsBest For
Llama 3.3 70B1Good open model
Llama 3.1 405B2Largest open model
Llama 3.1 70B1Balanced performance
Llama 3.1 8B0.5Quick and cheap

Meta strengths: Open weights, good value, fast inference

DeepSeek

ModelCreditsBest For
DeepSeek V3.21.5Reasoning and code
DeepSeek Chat V3.11General chat

DeepSeek strengths: Strong reasoning, competitive pricing

xAI (Grok)

ModelCreditsBest For
Grok 4.1 Fast2Quick responses
Grok 43Full capability
Grok 4 Fast1.5Balance of speed and quality

xAI strengths: Real-time knowledge, direct responses

How to Change Models

In Notebook Chat

  1. Look at the bottom of the chat input
  2. Click the model name (shows current model)
  3. Browse or search for a model
  4. Select to switch

In General Chat

Same process — click the model selector and choose.

Note

Model selection applies to future messages. Previous messages keep their original model.

Credit Costs Explained

Credits are consumed based on:

  1. Base model cost — Each model has a per-message rate
  2. Query complexity — Simple questions cost less than complex analysis
  3. RAG usage — Searching sources adds a small cost
  4. Web search — Optional, adds cost when enabled

Complexity Levels

LevelMultiplierExample Questions
Light1x”What is X?” / Simple lookups
Medium1.5x”Compare X and Y” / Analysis
Heavy2.5xMulti-step research / Deep synthesis

Example Costs

ScenarioBaseComplexityTotal
Simple question, Haiku0.51x~0.5 credits
Analysis question, Sonnet21.5x~3 credits
Deep research, Opus102.5x~25 credits
Tip

Start with a cheaper model for exploration. Switch to a more capable model for final, important questions.

Model Capabilities

Not all models support all features:

CapabilityDescriptionModels
ChatBasic conversationAll
Code GenerationWriting and debugging codeMost
Function CallingTool use and structured outputGPT-4+, Claude 3+
VisionUnderstanding imagesGPT-4o, Claude 3+, Gemini

Tips for Model Selection

Start Cheap, Scale Up

  1. Try your question with a fast, cheap model (Haiku, Llama 8B)
  2. If the response isn’t good enough, try a mid-tier model
  3. Reserve expensive models for truly complex tasks

Match Model to Task

Task TypeGood ChoiceWhy
SummarizingMid-tier (Sonnet, GPT-4o)Needs understanding, not deep reasoning
CodingSonnet, GPT-4oStrong at code
Complex reasoningOpus, o3/o4Designed for hard problems
Quick lookupsHaiku, Llama 8BSpeed over depth
Long documentsGemini ProLarge context window

Watch Your Credits

  • Check credit costs in the message metadata (info icon)
  • Monitor your balance in account settings
  • Set up alerts before running low

Learn More

Last updated on