IntelligenceLab/RSR-27B

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

IntelligenceLab/RSR-27B is a 27 billion parameter language model, finetuned from Qwen-3.8-27B using Recursive Self-Rewrite (RSR) trajectories. This model is specifically optimized for complex, hard terminal tasks, demonstrating improved efficiency and success rates compared to its base model. It excels in scenarios requiring iterative problem-solving and self-correction, such as data querying, systems infrastructure, and scientific simulations, by generating more concise and effective solution trajectories.

Loading preview...

RSR-27B: Recursive Self-Rewrite Model

RSR-27B is a 27 billion parameter model developed by IntelligenceLab, finetuned from Qwen-3.8-27B. Its core innovation lies in the application of Recursive Self-Rewrite (RSR) trajectories, coupled with additional Reinforcement Learning (RL) training to enhance stability on challenging terminal tasks. This approach allows the model to generate more efficient and successful solution paths compared to its base model.

Key Capabilities

  • Enhanced Problem Solving: Demonstrates significant improvements in solving complex, multi-step tasks across various domains.
  • Efficient Trajectory Generation: Capable of producing shorter, more direct solution trajectories, reducing the number of steps and agent time required.
  • Self-Correction and Refinement: Leverages RSR to iteratively refine its approach, leading to higher success rates on tasks where the base model might struggle or time out.
  • Domain Versatility: Proven effective in diverse areas including data querying (tree-sitter), systems & infrastructure (CMake), and physical science & simulation (OpenFOAM).

Good For

  • Automated Task Execution: Ideal for scenarios requiring autonomous agents to complete intricate, goal-oriented tasks.
  • Complex Code Generation & Debugging: Can be applied to tasks involving code generation, configuration, and debugging where iterative refinement is beneficial.
  • Scientific Simulation & Analysis: Suitable for automating workflows and problem-solving in scientific and engineering domains.
  • Improving Agent Efficiency: Developers looking to reduce computational steps and time for AI agents tackling difficult, multi-stage problems.

Finetunes
1 models