Models
Echo and Horizon
Model ids, context windows, thinking levels and how to choose.
Faelith serves two model families. Both run with a 1M-token context window, both accept text and images, and both stream.
| Model id | Family | Context | Best for |
|---|---|---|---|
echo | Echo | 1M | Fast edits, chat, everyday tasks |
horizon | Horizon | 1M | Multi-step reasoning, hard debugging, design |
The whole window is billed at the same per-token rate. Rates are on the Models page.
Thinking levels#
Both families support extended reasoning. The level controls how much of the budget the model may spend before answering:
| Level | Behaviour |
|---|---|
off | No extended reasoning; fastest |
low | Light reasoning for simple tasks (Echo default) |
medium | Skips thinking on simple queries |
high | Deep reasoning for complex tasks (Horizon default) |
xhigh | Always thinks thoroughly |
max | Maximum depth; slowest |
Set it with /thinking <level> in the CLI, the gauge picker in the app, or the reasoning_effort field on the API.
Model lineage#
Echo and Horizon are post-trained by Faelith on top of the DeepSeek V4 Flash and DeepSeek V4 Pro open-weight base models respectively, then served on Faelith infrastructure with Faelith prompts, tools and safety layers. Behaviour, pricing and data handling are Faelith's; upstream model releases do not change a model id under you.
Found a mistake or a gap? Tell us and we will fix the page.