saturday-labs/turkish-banking-agent-1.5b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 17, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The saturday-labs/turkish-banking-agent-1.5b is a 1.5 billion parameter model developed by saturday-labs, fine-tuned for function calling in Turkish banking scenarios. Built upon Qwen/Qwen2.5-1.5B-Instruct, it translates natural Turkish user requests into appropriate tool calls and arguments, then summarizes tool results in Turkish. With a 32768 token context length, it excels at handling single and multiple tool calls, clarifying missing information, and requesting confirmation for destructive operations within a banking context.

Loading preview...

Turkish Banking Agent 1.5B Overview

The saturday-labs/turkish-banking-agent-1.5b is a specialized 1.5 billion parameter model developed by saturday-labs, designed for function calling in Turkish banking contexts. It is built on Qwen/Qwen2.5-1.5B-Instruct and further trained using LoRA, with continued training on a 4096 token context window. The model's primary function is to interpret natural Turkish user requests, convert them into suitable tool calls with arguments, and then summarize the tool's output back to the user in Turkish.

Key Capabilities

  • Tool Calling: Supports both single and multiple tool calls.
  • Clarification: Proactively asks the user for missing mandatory arguments instead of guessing.
  • Confirmation: Requests user confirmation for destructive operations like transfers or card cancellations.
  • Result Summarization: Converts tool results into coherent Turkish responses.
  • Scope Management: Avoids unnecessary tool calls for requests that don't require them and rejects out-of-scope requests.

Performance Highlights

Evaluated on unseen tools within a held-out test set of 598 examples, the model demonstrates strong generalization capabilities:

  • Tool Calling Decision Accuracy: 96.3%
  • Full Tool Name Accuracy: 87.9%
  • Full Call Accuracy: 83.0%
  • JSON Validity: 97.0%

Good for

  • Turkish Banking Applications: Ideal for integrating natural language interfaces into Turkish banking systems.
  • Function Calling: Excels at translating user intent into structured API calls.
  • Prototyping & Research: Suitable for developing and testing conversational AI agents in financial domains.

Limitations

It's important to note that the model is released for research and prototyping. Production use requires handling certain aspects at the application layer, such as validating numerical values, confirming tool names, managing complex multi-step scenarios, and providing external knowledge for banking information queries (as the model is not a knowledge base).

Finetunes
1 models