Guides
Choosing a Model
Pick the right model for your translation needs across different providers.
The model you use affects translation quality, speed, and cost. Here's what to consider for each provider type.
Local Models (LM Studio / Ollama)
Recommended: HY-MT1.5-7B
HY-MT1.5-7B is the recommended starting point for LM Studio. It balances
quality and performance across Karpa's supported languages and runs comfortably
on consumer hardware. The Q4_K_M quantized variant is a good pick.
Factors to consider
| Factor | Guidance |
|---|---|
| Quality | Larger / less-quantized models translate better. |
| Speed | Smaller, more-quantized models are faster. |
| Memory | Match model size to your available RAM/VRAM. |
| Languages | Some models are tuned for specific language pairs. |
Quantization explained
Quantization compresses a model to use less memory, with a small quality trade-off.
| Level | Trade-off |
|---|---|
Q4_K_M | Sweet spot — good quality, low memory. |
Q5_K_M | Slightly better quality, a bit larger. |
Q8_0 | Near full precision, largest footprint. |
Cloud Models (OpenAI / Anthropic / Gemini / OpenRouter)
| Provider | Default Model | Good For |
|---|---|---|
| OpenAI | gpt-4o-mini | Fast, affordable, great quality |
| Anthropic | claude-sonnet-4-20250514 | Nuanced translations, longer context |
| Gemini | gemini-2.0-flash | Fast, free tier available |
| OpenRouter | amazon/nova-2-lite-v1 | Wide model selection, competitive pricing |
Trying a new model
- For local models: Download the model in LM Studio / Ollama, then load it.
- For cloud models: Just select from the dropdown in Settings.
- Update the model in Settings → Connection.
- Test with a translation.
If translations go wrong after a model swap, the model name probably doesn't match — double-check for typos.