Choosing the Right AI Model
This guide helps you pick the right model for your command and workflow.
Practical Baseline
RewriteBar is tested with a variety of models from different providers. But I personally use OpenAI for daily use.
- Start with GPT-5.2 for most writing tasks.
- You can also try leading Anthropic models for high-quality rewriting and reasoning-heavy prompts.
How to Choose by Task
Fast, repetitive edits
Use faster and cheaper models for:
- Grammar and spelling fixes
- Case changes
- Simple tone changes
Complex rewriting and prompt work
Use stronger models for:
- Prompt improvement
- High-stakes copy
- Nuanced tone adaptation
- Structure-heavy transformations
Privacy-first or offline workflows
Use local providers (Ollama / LM Studio) when privacy or offline usage is your priority.
Local Models
Tested and Recommended
The general recommendation is to start with models in the 3B range, like llama 3.2 3b as a good starting point for most users. You can also use OpenRouter to try models quickly without installing anything.
| Model | Quality | Comments |
|---|---|---|
| llama 3.2 3b | ⭐⭐⭐⭐⭐ | Best starting point for most users |
| gpt-oss 20b | ⭐⭐⭐ | Good but requires more RAM |
TODO: Table of models
Tested but Weak Experience
Models in this section run, but the quality is not good enough for reliable daily use. This means you may experience issues such as results in different languages, explanatory text, or just bad results.
- Gemma 3 1b: This model is too small for most tasks.
- Gemma 3 4b: This model does not follow the instructions well.
Command-Level Overrides
A common setup:
- Pick one global default model for daily commands.
- Override model per command when quality requirements are higher.
Example:
- Keep a fast default for grammar/spelling.
- Use a stronger model for prompt-improvement commands.
Decision Checklist
- Is this command low-risk and repetitive? Prefer a faster model.
- Does output quality matter more than speed? Prefer a stronger model.
- Does text include sensitive content? Consider a local model.
- Is this command used often? Set a command-level model override if needed.
