OTA-AI/OTA-v1

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 4, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

OTA-v1 is a 14.8 billion parameter Browser Agent Model (BAM) developed by OTA-AI, fine-tuned from the Qwen2.5 base. It is specifically designed for controlling browser environments and performing automated web tasks using frameworks like browser-use. This model excels in multi-step planning, precision tool utilization, and long-context processing of full-page DOM structures up to 128K tokens. It is optimized for cost-efficient deployment on consumer-grade GPUs, enabling local execution for web automation and interaction.

Loading preview...

OTA-v1: A Specialized Browser Agent Model

OTA-v1 is a 14.8 billion parameter Browser Agent Model (BAM) developed by OTA-AI, fine-tuned from the Qwen2.5 base architecture. Unlike traditional instruction-tuned models, OTA-v1 is specifically optimized for reasoning and tool use within browser contexts, making it highly effective for web automation and interaction.

Key Capabilities

  • Cost-Efficient Deployment: Optimized for consumer-grade GPUs (NVIDIA 3090/4090) with 16-bit precision (20GB VRAM) and 4-bit quantization (10GB VRAM), supporting local execution.
  • Multi-step Planning Engine: Automatically breaks down complex tasks into executable action sequences, incorporating conditional logic for error recovery and maintaining state awareness across browser sessions.
  • Precision Tool Utilization: Offers native support for browser agent frameworks like browser-use, with automatic detection of interactive elements and form fields.
  • Long-Context Optimization: Capable of processing full-page DOM structures up to 128K tokens, utilizing YARN-enhanced attention patterns for efficient HTML traversal and context-aware element resolution.
  • Structured Execution: Generates robust tool use instructions, including formatted output under long context and self-correction based on action history.

Good For

  • Automated web tasks and browser control.
  • Complex multi-step interactions within web environments.
  • Applications requiring precise tool utilization in dynamic web applications.
  • Local deployment of browser automation agents on consumer hardware.

Finetunes
1 models