Models
The Talos model id and how reasoning modes work
The Ablatic API serves one model, Talos m1, with the id talos-preview (talos is accepted as an alias in requests). This page explains how to pick a reasoning mode.
The model
| Model | Modes | Context window | Modalities |
|---|---|---|---|
talos-preview | fast, thinking | 200000 | multimodal input, text output |
You do not switch models to change how hard Talos thinks. talos runs in one of two reasoning modes — fast or thinking — and within thinking you set the depth per request.
How mode selection works
On the chat endpoint, the reasoning_effort parameter controls the mode and the reasoning depth. It accepts the standard OpenAI values:
reasoning_effort | Mode | Reasoning depth |
|---|---|---|
none, minimal | fast | no reasoning block |
low | thinking | brief — the model is asked to keep its reasoning short |
medium (default) | thinking | default |
high, xhigh, max | thinking | larger reasoning-token budget |
When you omit reasoning_effort, the default is medium (thinking). Values outside the list fall back to thinking rather than failing, so a client that sends an unusual effort string still gets a valid response. Above medium the effort raises the reasoning-token ceiling; it does not switch to a different mode.
On the Anthropic-style endpoint (/v1/messages), the Anthropic model name maps by substring: haiku to fast, sonnet and opus to thinking, with opus given the largest reasoning-token budget, so an unmodified Anthropic app runs unchanged.
Modalities
Talos accepts multimodal input: text, image, and video. It returns text.
Listing models
GET /v1/models returns the single talos-preview model and needs no auth. See the Models endpoint reference.