Docs
Capabilities

Text generation

Generate chat completions with Talos m1

Send a list of messages and get a text reply with POST /v1/chat/completions.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.ablatic.ai/v1",
    api_key="sk-ablatic-...",
)

resp = client.chat.completions.create(
    model="talos",
    messages=[
        {"role": "system", "content": "You are a concise assistant."},
        {"role": "user", "content": "Name three rivers in Austria."},
    ],
    temperature=0.7,
    max_tokens=512,
)
print(resp.choices[0].message.content)

Messages

The messages array carries the conversation. Each item has a role and content. Three roles are used:

  • system sets behavior and instructions for the model.
  • user is the human turn.
  • assistant is a prior model turn, which you include to continue a conversation.

You can send 1 to 200 messages. The full context window is 200000 tokens.

Parameters

ParameterTypeNotes
modelstringDefaults to talos.
messagesarrayRequired. 1 to 200 items.
temperaturenumber0 to 2. Higher is more random.
max_tokensinteger1 to 200000. Caps the generated reply.
top_pnumberNucleus sampling.
reasoning_effortstringnone/minimal (fast), low (brief thinking), medium (thinking), high/xhigh/max (thinking, larger reasoning budget).

Reasoning effort

talos runs in thinking mode by default. The reasoning_effort field controls how much the model reasons before answering: none and minimal map to fast, low to brief thinking, medium to thinking, and high (and xhigh/max) raise the reasoning-token budget. When you omit reasoning_effort, the request runs in thinking mode, not the fastest path.

For the lowest latency, set reasoning_effort to none, which selects fast mode.

resp = client.chat.completions.create(
    model="talos",
    reasoning_effort="none",
    messages=[{"role": "user", "content": "Give me a one-line summary."}],
)

Thinking consumes output tokens, so set max_tokens high enough for the reasoning to finish. Too small a value fails with an error rather than returning a partial answer. Use a large max_tokens (1024 or more) for high effort (high/xhigh/max), which reasons longer and runs slower. See Reasoning.

Response

The response is a chat.completion object. The reply text is in choices[0].message.content. The usage object reports prompt_tokens, completion_tokens, and total_tokens. A reasoning_tokens count appears when the model reasons.

Next