tase-8b

tase-8b is a compact model trained by Tase Industries for one job: driving the Tase app. It is specialized for tool calling — deciding which tool a request needs and producing correct arguments for it — and it does this with low latency in both Turkish and English.

What it is designed for

• Tool selection: mapping a natural-language request to the right Tase tool, even when the request is short, colloquial, or mixes Turkish and English.
• Argument generation: producing well-formed, complete arguments for the selected tool — dates, amounts, titles, and identifiers extracted from the request.
• Low-latency chat: short conversational turns where response speed matters more than long-form reasoning.

Why tool calling is fast

Most APIs require you to send the full tool schema with every request: names, descriptions, parameter types, all of it, on every call. tase-8b works differently. The Tase tool schemas are trained directly into the model weights, so the model already knows every tool it can call and the exact shape of each one.

• No schema payload per request — requests carry only the conversation, so they are significantly smaller.
• No schema parsing at inference time — the model does not spend tokens re-reading tool definitions it already knows.
• Fewer malformed calls — argument shapes are learned, not inferred from a JSON blob on the fly.

Capabilities

• Chat completions — standard multi-turn conversation.
• Function calling — built-in Tase tools, invoked without sending schemas.
• Streaming — token-by-token responses over server-sent events.

Limits

tase-8b is compact by design. For requests that require deep reasoning over long context — multi-step planning, long-document analysis, involved multi-constraint problems — escalation to a larger model is recommended. In practice you rarely manage this yourself: routing detects these requests and escalates automatically. See Routing on the Models page.

Example request

A tool-calling request. Note that no tool definitions are sent — the model already knows the Tase tools.

curl https://api.tase.app/v1/chat/completions \
  -H "Authorization: Bearer $TASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tase-8b",
    "messages": [
      { "role": "user", "content": "Add 40 dollars for groceries to today" }
    ]
  }'

{
  "id": "cmpl_9f2c1a",
  "object": "chat.completion",
  "model": "tase-8b",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "tool_calls": [
          {
            "id": "call_51xk8d",
            "type": "function",
            "function": {
              "name": "add_expense",
              "arguments": "{\"amount\": 40, \"currency\": \"USD\", \"category\": \"groceries\", \"date\": \"today\"}"
            }
          }
        ]
      },
      "finish_reason": "tool_calls"
    }
  ]
}