Skip to main content
Search

# Qwen3.6-35B-A3B

5 related articles

Subscribe to updates
  1. goodfire.com

    Goodfire reads Qwen3.6 neurons directly: ablating 92.2% of weak activations raises language-model loss by only 2.9%

    Goodfire studied the MLP neurons of Qwen3.6-35B-A3B directly instead of first learning a replacement feature space. Keeping only the 320 strongest router-weighted neurons per forward pass and ablating the other 92.2% raised language-model loss by 2.9%; most neurons carry one computational mode or one dominant mode, and natural-language descriptions of them can predict, simulate and drive their firing.

    goodfire.com
    Neuron InterpretabilityQwen3.6-35B-A3BActivation SparsificationNeuron Subcircuits and ModesNeuron Scientist
  2. arXiv

    Terminal agents build their own RL environments: up to 25x more valid ones, 75% cheaper each

    AutoEnvScaling turns reinforcement-learning environment design into a terminal task, letting a proposer agent build new environments from the solver's rollouts, raising the valid-environment rate by up to 25 times and cutting cost per valid environment by up to 75%.

    University of WashingtonMicrosoft (United States)Stanford UniversityUniversity of North Carolina at Chapel HillNortheastern University
    Simon Yu（First author）
    AutoEnvScalingRecursive Self-ImprovementTerminal agentsReinforcement LearningQwen3.6-35B-A3B
  3. arXiv

    Splitting document QA into a searcher and a reader: same model gains 11.0 F1 on M3DocVQA

    Open-domain multimodal document QA requires first locating the right passages and then reading them, yet current retrieval-augmented agents give one model both kinds of tools and let it schedule itself. KnowForge separates the pipeline by job: a search-only Locator agent, a code-scheduled bridge judge that forces the missing second hop, a bounded evidence pool merging every search ranking, and a Deep Reader that only reads text, images and tables. With a self-hosted Qwen3.6-35B-A3B, it reaches F1 47.5 on the full M3DocVQA (2,441 questions, 3,368 PDFs), 11.0 above the published M3DocRAG and 7.1 above one-shot retrieve-and-answer on the same index and model.

    Wenlin Lin（First author）
    KnowForgehybrid-retrieval knowledge-point indexQuestion AnsweringQwen3.6-35B-A3B
  4. arXiv

    ROSS reuses discarded historical self-generated rollouts, lifting Qwen3.6-35B-A3B's six-benchmark MOPD average from 58.40% to 62.20% and SWE-bench Verified from 64.20% to 68.40%

    ROSS introduces a selective-supervision relearning procedure that takes historical self-generated rollouts saved from domain-specific RL, multi-teacher on-policy distillation (MOPD), or agentic RL, uses an outcome verifier to keep successful trajectories and an LLM reviewer to mark which model-generated spans are worth imitating, then applies loss only to those selected tokens while keeping the full trajectory as context, improving upstream checkpoints through an offline SFT stage without new policy rollouts and raising Qwen3.6-35B-A3B's six-benchmark MOPD average from 58.40% to 62.20% and SWE-bench Verified from 64.20% to 68.40%.

    Zhiwei Zhang（First author）
    Selective SupervisionLLM Post-TrainingOn-Policy DistillationAgentic Reinforcement LearningQwen3.6-35B-A3B
  5. arXiv

    CompoWorld composes 448 reusable services into cross-service tasks, lifting Qwen3.6-35B-A3B by 9.17 points on average across eight benchmarks

    CompoWorld introduces compositional environment scaling: coding agents turn MCP tool specifications into executable services with typed states and shared interfaces, a random walk connects services through dependency graphs to generate and verify cross-service tasks, verified trajectories support SFT while a Completion-Focused Rubric Reward guides GRPO-based RL, and with 448 services exposing 10,130 tools plus 3K SFT trajectories and 1K RL tasks, Qwen3.6-35B-A3B improves by 9.17 points on average across eight benchmarks and raises AutomationBench task success from 10.33% to 32.33%.

    Xiao-Wen Yang（First author）
    CompoWorldCompletion-Focused Rubric RewardGRPOSupervised Fine-tuningWorld Model

You’re all caught up