sabicap/sabi-noema

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 13, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

Sabi-Noema is a 27 billion parameter agentic language model developed by sabicap, built upon Qwen3.8-27B and fine-tuned using open-source RL environments. It excels in agentic tasks, software engineering, and tool calling, demonstrating significantly reduced hallucination rates (15.1%) and improved token efficiency compared to its base model. This model is optimized for long, multi-action jobs, customer-facing agents requiring policy adherence, and cost-sensitive agentic deployments.

Loading preview...

Sabi-Noema: An Agentic Model for Enhanced Reasoning and Efficiency

Sabi-Noema is a 27 billion parameter agentic model developed by sabicap, built on the Qwen3.8-27B architecture. It has been exclusively trained using open-source Reinforcement Learning (RL) environments to push the boundaries of small model capability and reasoning efficiency, particularly in agentic workflows.

Key Capabilities

  • Superior Agentic Performance: Achieves 76.0% on Terminal-Bench 2.1 and 47.5% on DeepSWE, outperforming its base model and several other benchmarks.
  • Reduced Hallucination: Demonstrates a significantly lower hallucination rate of 15.1% on AA-Omniscience, compared to 30.4% for Qwen3.8-27B.
  • High Token Efficiency: Uses 2.3 times fewer output tokens than Qwen3.8-27B and 2.0 times fewer than Swift 1.5 on Terminal-Bench tasks, leading to more cost-effective agentic serving.
  • Robust Tool Calling: Shows strong performance in tool-calling benchmarks like BFCL v4 multi-turn (63.9%) and MCPMark (48).
  • Long Context Handling: Supports a substantial context length of 262,000 tokens, suitable for complex, multi-turn interactions.

Good For

  • Long-running agentic jobs: Ideal for tasks requiring many actions and end-of-task verification in shell environments or with external tools.
  • Customer-facing agents: Suitable for agents that need to adhere to policies, handle difficult customer interactions, and make correct tool calls over extended conversations.
  • Applications requiring reliability: Beneficial for assistants that need to explicitly state when they lack information, due to its low hallucination rate.
  • Cost-sensitive deployments: Its high token efficiency makes it a strong candidate for agentic serving where output token costs are a concern.
  • High-density GPU serving: The related Sabi-Noema-MLA model offers optimized attention cache for running many long sessions on a single GPU.

Finetunes
2 models