frontier-infra/jebadiah-9b-v2

Hugging Face
VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 25, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

Jebadiah 9B v2 is a 9 billion parameter open System One style decision model developed by Frontier Infra, built on Qwen/Qwen3.5-9B. Unlike traditional LLMs, it answers typed questions (choice, noul, score) with a probability distribution over option labels in a single forward pass, rather than generating text. This model is optimized for structured decision-making tasks, providing calibrated probabilities for classification and scoring, and is compatible with TypeSafe-compatible /v1/systemone endpoints.

Loading preview...

Jebadiah 9B v2: A Specialized Decision Model

Jebadiah 9B v2, developed by Frontier Infra, is a 9 billion parameter model designed for structured decision-making rather than text generation. Built upon the Qwen/Qwen3.5-9B base, it operates as an "open System One style decision model," providing probabilistic answers to specific question types in a single forward pass.

Key Capabilities

  • Typed Question Answering: Handles three distinct question types:
    • Choice: Selects one option from N possibilities.
    • Noul: Evaluates a yes/no statement, returning P(yes).
    • Score: Places a state on an ordered rubric.
  • Probabilistic Outputs: Returns a probability distribution over option labels, offering calibrated confidence for each decision.
  • API Compatibility: Serves a TypeSafe-compatible /v1/systemone endpoint, allowing integration with existing Jev clients.
  • Standard Transformers Model: Can be run in various environments, including transformers, vLLM, llama.cpp (via GGUF), and MLX for Apple silicon.
  • Performance: Achieves a JevBench v1.4.2 score of 0.818 on public items and a Headline accuracy of 73.9% across various evaluation sets.

Good For

  • Automated Decision Systems: Ideal for applications requiring structured, probabilistic decisions, such as ticket routing, urgency assessment, or content moderation.
  • Classification Tasks: When a clear, probabilistic classification is needed over open-ended text generation.
  • Integration with Existing Decision Engines: Seamlessly integrates with systems expecting System One style outputs.
  • Resource-Constrained Environments: The 9B size offers a balance of performance and efficiency, with GGUF and MLX versions available for local deployment.

Finetunes
10 models