runelab-ai/Hati-Glint-SQL-9B-v1

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Hati-Glint-SQL-9B-v1 is a 9-billion parameter text-to-SQL model developed by RuneLab.ai, fine-tuned from Qwen3.5-9B with a 32768-token context length. It specializes in converting Turkish natural language questions into executable SQLite SQL queries, achieving 73.21% execution accuracy on TURSpider and 47.26% on BIRDTurk in direct mode. The model also functions as a general Turkish assistant and demonstrates improved tool-calling accuracy and token efficiency, making it suitable for on-premise AI agents and business intelligence applications requiring Turkish language support.

Loading preview...

Hati Glint SQL: Turkish Text-to-SQL Specialist

Hati Glint SQL (Hati-Glint-SQL-9B-v1) is a 9-billion parameter, open-weight model developed by RuneLab.ai, fine-tuned from Qwen3.5-9B. It excels at converting Turkish natural language questions into executable SQLite SQL queries, making it the strongest open Turkish text-to-SQL model under 10B in RuneLab.ai's evaluation. Beyond SQL, it retains the strong general Turkish capabilities of its base model and shows improved tool-calling accuracy.

Key Capabilities

  • High Accuracy Turkish Text-to-SQL: Achieves 73.21% execution accuracy on TURSpider and 47.26% on BIRDTurk in direct mode, outperforming its base model by +9.1 and +7.2 points respectively. With thinking mode and agent feedback, scores reach 77.85% on TURSpider and 55.35% on BIRDTurk.
  • Token Efficiency: In thinking mode, it uses approximately 3-3.6 times fewer reasoning tokens than the base model, leading to faster responses and reduced computational cost.
  • Enhanced Tool Calling: Shows a +2.0 exact-call accuracy improvement on Turkish tool calls from unseen tool families, crucial for business process automation.
  • General Turkish and English Proficiency: Maintains the strong general language abilities of Qwen3.5-9B, with a notable +4.8 improvement on English Spider text-to-SQL.
  • Agent Mode with Execution Feedback: Features an agent wrapper that allows the model to correct SQL errors based on execution feedback, significantly boosting accuracy with minimal token overhead.

Good For

  • Turkish Natural Language Queries: Ideal for asking questions over company databases in Turkish for BI, reporting, and analytics.
  • On-Premise Deployments: Suitable for scenarios where data privacy (e.g., KVKK compliance) requires processing data within a company's own infrastructure on a single GPU.
  • AI Agents: Functions as a versatile component for AI agents that need to chat in Turkish, generate SQL, and call tools within a single model.
  • Business Process Automation: Effective for tool and function calling in ERP, CRM, finance, HR, and support systems.

Finetunes
1 models