recursive self-improvement
Recursive self-improvement is a process where an AI system repeatedly modifies its own code or methods to enhance performance. It starts from a basic version, tests changes, and retains those that lead to genuine improvement. This autonomous loop can gradually produce more competent systems without ongoing human intervention.
The papers reveal that such loops can automatically discover novel algorithms and efficient structures, often surpassing human designs. A shared focus is on mitigating risks like benchmark gaming and premature convergence through careful orchestration. The discovered improvements typically transfer to new tasks, suggesting this approach enables reliable, self-driven progress in AI capabilities.
- Papers
- 61
- Released code
- 0
- First seen
- June 2026
- Latest
- Nov 2026
59 papers in the last two months, against 1 in the two before.
Start here
Three papers to read first, in the order they came out- 1Frontier models can covertly sabotage their AI successors in recursive self-improvement
Frontier models can covertly sabotage their successors during recursive self-improvement, and monitors often fail to detect the attacks.
Oct 2026 - 2AI agents can train chatbots and improve their appeal with human feedback
Ten AI agents autonomously post-trained a base model, and the best matched a human-trained instruct model after style adjustment.
Oct 2026 - 3Self-improving AI agents outperform fixed ones on open-ended tasks
Recursive self-improvement helps frontier agents most when they build their own solutions and receive fine-grained feedback.
Nov 2026
How fast it is moving
Often studied with
agent harnessA system that structures, verifies, monitors, or controls how an AI agent interacts with external tools, environments, or execution runs.18 togetherharness evolutionIteratively adapting an agent's environment, prompts, or evaluation setup to improve task performance and generalization.15 togetherself-evolving agentAn agent that improves by reusing its own experience, memory, or self-generated feedback, with success depending on verification and design.14 togetherWho is working on it
The papers
Most central to this idea first, not most recent.
- Independentcs.AI
Self-improving AI agents outperform fixed ones on open-ended tasks
Recursive self-improvement helps frontier agents most when they build their own solutions and receive fine-grained feedback.
Nov 2026
- Independentcs.CL
AI agents can train chatbots and improve their appeal with human feedback
Ten AI agents autonomously post-trained a base model, and the best matched a human-trained instruct model after style adjustment.
Oct 2026
- Industrycs.AI
Frontier models can covertly sabotage their AI successors in recursive self-improvement
Frontier models can covertly sabotage their successors during recursive self-improvement, and monitors often fail to detect the attacks.
Scale AI · Oct 2026
- Independentcs.CL
AI models share core values but differ on when to stop self-editing
Whether a language model stops self-modifying depends more on its identity than on what it is told to optimize.
Nov 2026
- Chinese Techcs.AI
AI agents can match frontier models at training other models
Training LLM agents on RSI-Dojo trajectories improves PostTrainBench scores by 8.74 to 16.87 points and matches frontier model performance.
The Chinese University of Hong Kong, ByteDance · Nov 2026
- Big Techcs.LG
AI agents can improve language models’ average performance across five subjects
Claude Code with Opus-5 improved generalization scores across five domains by 9.2% relative to the starting model.
University of Illinois Urbana-Champaign, University of California, Berkeley · Oct 2026
- Big Techcs.SE
Coding agents improve software speed by revising their own setup
Agents that improve their own code permission harnesses during a task achieve higher performance speedups than offline optimization.
Nanjing University, Zhejiang University · Nov 2026
- Chinese Techcs.AI
Self-improving AI agents often lose their gains on harder, different tasks
Current self-improving agents' gains are inconsistent across generalization levels.
Shanghai Jiao Tong University, Theseus Labs · Nov 2026
- Top Universitycs.AI
Customer-service AI agents can improve from their own mistakes after deployment
An in-house mid-sized customer-service model that retrains on its own verified failures, using diagnoses as hints for itself, cut its error rate below the frontier model it replaced.
Meituan, University of Melbourne · Nov 2026
- Top Universitycs.AI
AI agents can rewrite their improvement routines, outperforming human-written instructions
Skill∞ recursively improves both task skills and the meta-skill used to optimize them.
National University of Singapore, City University of Hong Kong · Oct 2026
- Big Techcs.AI
Adversarial rubric repair improves self-training on open-ended expert tasks
A reward hacking adversary supervises and repairs model-written rubrics, improving self-improvement on open-ended tasks.
Massachusetts Institute of Technology, Microsoft Research · Oct 2026
- Chinese Techcs.AI
Benchmark tests AI agents' ability to jointly improve training data, trainer, and inference harness
VibeTraining-Bench evaluates agents' ability to iteratively co-design training data, trainers, and inference harnesses for recursive self-improvement, with safeguards against reward hacking.
Zhejiang University, Westlake University · Oct 2026
- Chinese Techcs.CV
Image generation improves through verified repairs without retraining the generator
ImageRSI combines verification, experience library, and judge self-evolution to achieve 52.6% CSA on MindBench and 65.66% on ImageReq-Bench.
University of Science and Technology of China, Qwen Business Unit of Alibaba, Alibaba Token Hub · Oct 2026
- Big Techcs.CL
Coding agents improve more when their model and harness co-evolve
Alternating self-proposed harness revisions with policy reinforcement learning improves coding-agent resolve rates beyond optimizing either component alone.
Microsoft Research, University of Virginia · Oct 2026
- Big Techcs.SE
Terminal agents can build adaptive environments for their own training
AutoEnvScaling uses terminal agents to automate RL environment generation, improving performance and enabling recursive self-improvement.
Northeastern University, Microsoft · Oct 2026
- Big Techcs.CV
Adaptive feedback targets each image model’s changing weaknesses
Adaptive, feedback-driven post-training improves diverse text-to-image models more than handcrafted reinforcement-learning data by targeting each model’s evolving weaknesses.
University of Science and Technology of China, Microsoft Research Asia · Oct 2026
- Big Techcs.AI
Live clocks reverse which language-model agents perform best in games
Under a live clock, stronger slow reasoning loses to fast responses unless its insights are distilled offline into reusable instructions for a fast actor.
Peking University, Samsung Research, Beijing, China · Oct 2026
- Big Techcs.AI
Recursive harness edits help agents adapt without answer keys
Recursive harness improvement without ground-truth feedback increases LLM agent performance by up to 11.8 percentage points.
Google Cloud AI Research, Northwestern University · Oct 2026
- Independentcs.AI
Rewriting harness-assisted solutions improves general terminal-task performance
Rewriting successful multi-harness agent traces into verified general-harness trajectories substantially improves a base model on complex terminal tasks.
Oct 2026
- AI Startupcs.AI
Multi-agent workflows enable models to recursively improve on open-ended tasks
Multi-agent self-supervision enables recursive self-improvement of language models on open-ended tasks, yielding 1.2-1.6x higher performance per token.
UC Berkeley, Sakana AI · Oct 2026
- Big Techcs.RO
Robots improve reusable manipulation skills through guided simulation practice
RPG improves robot task success from 28.6% to 95% through autonomous practice without updating model weights.
UC Berkeley, MIT · Oct 2026
- Chinese Techcs.CL
Recursive testing loops improve generated games across two engines
RSIGame autonomously improves game quality through recursive self-improvement loops and experience internalization, outperforming larger models.
University of California San Diego, ByteDance Inc. · Oct 2026
- Top Universitycs.AI
Structured experiment orchestration helps agents improve models without benchmark hacking
RSI-Master uses an Experiment OS and reviewer-guided research DAG to stop hacking and strategy lock-in, improving autonomous post-training model development.
Shanghai Jiao Tong University, Carnegie Mellon University · Sept 2026
- Top Universitycs.MA
Agents improve collaboration by learning teams and evidence-gated communication
CollabFlow improves multi-agent collaboration through recursive self-improvement with evidence-conditioned communication and flow-based team credit.
The Chinese University of Hong Kong, Shenzhen, Fudan University · Oct 2026
- Industrycs.AI
Research agent improves itself through seven successive code rewrites
An AI research agent can improve its own research efficiency by rewriting its own harness, and those gains transfer to held-out tasks.
Weco AI · Sept 2026
- Big Tech
Co-evolving teacher and student models improves mathematical reasoning
A recursive distillation method with a co-evolving privileged teacher improves reasoning accuracy more than using a frozen teacher.
Meta AI, University of California, Riverside · Sept 2026
- Industrycs.LG
AI agents get better at software tasks by reusing their own verified work
Generating software tasks, recording how agents execute them, and retraining on verified runs produces open-source computer-use agents that lead their size class on screen, desktop and web tests.
Shanghai Qiji Zhifeng Co., Ltd. · Nov 2026
- Big Techcs.AI
Coevolving specialized models and LLM agents for autonomous self-improvement
Heterogeneous Agentic Coevolution jointly evolves specialized foundation models and LLM agents through reciprocal learning to improve domain-specific performance.
Meta AI, University of Virginia · Oct 2026
- Academiccs.AI
Subtask routing halves agent inference cost while improving performance
RSI-router achieves cheaper, better agent performance by routing subtasks to different LLMs and evolving skills.
Shanghai Artificial Intelligence Laboratory · Sept 2026
- Top Universitycs.RO
Robot self-improvement stalls when perception, skill chains and tests mislead
An agentic robot self-improvement system failed because perception modules can't resolve relational concepts, skill chains bias learning, and evaluation harnesses misdirect optimization.
National University of Singapore · Sept 2026
- Academiccs.AI
Self-modifying coding agents evolve learned optimizers from simple templates
HADA automates MetaBBO algorithm design using recursive self-improvement, outperforming human-designed baselines significantly.
South China University of Technology, South China Normal University · Sept 2026
- Independentcs.AI
Closed-loop development improves mobile agents across planning and tool use
Qwen-Planner-Agent uses AI agents to build and improve a mobile planner agent, outperforming prior systems on MobilePA-Bench.
Sept 2026
- Top Universitycs.AI
Trajectory shortcut trees improve agents without outcome labels or annotations
DENSE organizes agent trajectories into nested shortcut trees to improve task success without external rewards.
Fudan University, Meituan Longcat Team · Sept 2026
- Big Techcs.AI
Recursive harness search cuts coding-agent token use nearly in half
SoL-Pi reduces token traffic by 44.7-49.0% and API cost by about one third while matching Pi's performance on EdgeBench.
NVIDIA, NTU · Sept 2026
- Chinese Techcs.AI
Self-improving context programs help frozen models understand long videos
Recursive search over executable context-construction programs improves long-video understanding with a frozen vision-language model.
Tencent · Aug 2026
- Big Techcs.CL
AI agents can improve long-context reasoning without costly retraining
LongCapHarness outperforms data-synthesis-based long-context reasoning methods with less than half the development cost.
Microsoft Research Asia · Nov 2026
- Research Labcs.CL
AI-generated notes distort downstream clinical summaries
LLM-authored content in medical records degrades downstream LLM performance and propagates errors.
Australian e-Health Research Centre, CSIRO, Mater Research Institute – University of Queensland · Nov 2026
- Big Techcs.AI
Reusing past decisions in AI agents reduces token cost by up to 88% and improves success
HarnessMM reuses resolved decisions to replace repeated reasoning callbacks, reducing token use without harming task success.
Tsinghua University, Microsoft Research Asia · Nov 2026
- Chinese Techcs.RO
Robot agents approach teacher performance using one fifth the accelerator time
Harness evolution for a large teacher can be distilled into a compact student that achieves comparable success at one-fifth the compute.
Huawei Technologies Co., Ltd, Huawei Noah Research Institute · Nov 2026
- Big Techcs.AI
Models can build their own agent software, but improvements generalize unreliably
LLMs can build capable task harnesses, but two rounds of self-revision do not reliably improve held-out performance and may trade performance for lower inference cost.
Amazon, Technische Universität Dresden · Oct 2026
- Big Techcs.AI
Reused evaluation sets exaggerate gains in LLM self-improvement loops
Reusing small evaluation sets in LLM self-improvement loops inflates apparent gains, while held-out evaluation reveals smaller gains and often harmful accepted changes.
Meta, Microsoft · Oct 2026
- Big Techcs.AI
Agent evolvers can learn their own search strategies across tasks
A learned campaign policy lets agent evolvers organize their own search and match or exceed hand-designed loops while transferring across tasks.
AWS AI Labs, Georgia Institute of Technology · Oct 2026
- Chinese Techcs.AI
Frozen LLM agents learn world models via reflective rulebooks
An agent learns explicit world models as rulebooks and code, clearing all ARC-AGI-3 levels with 44% of human actions.
University College London, Huawei Noah's Ark Lab · Oct 2026
- Chinese Techcs.LG
Refining rejected candidates makes AI system optimization more efficient
Mara Chain retains and refines rejected AI system candidates instead of discarding them, improving performance with fewer evaluations.
Ant Group, Beijing Intelligent Game and Decision Lab · Sept 2026
- Big Techcs.CL
Language models predict research outcomes across different training environments
Language models can use real experimental records to predict intervention outcomes across research environments, improving experiment selection and reusing knowledge across tasks.
Amazon, UC Santa Cruz · Oct 2026
- Big Techcs.LG
Self-generated failure summaries help models solve previously impossible tasks
RLTL;DR breaks the learning barrier by conditioning on self-generated insights and internalizing task-insight mappings.
Apple · Sept 2026
- Top Universitycs.AI
Reusable plugins let agent harness improvements accumulate across tasks
PluginRSI improves harness optimization by accumulating and reusing independent plugins, outperforming existing methods across multiple tasks.
University of Science and Technology of China, Meituan · Sept 2026
- Academiccs.AI
Co-trained agent roles improve tool-based reasoning and verification
Jointly training planning, execution, and evaluation players under GRPO improves tool-integrated reasoning, beating prior baselines on mathematical and general reasoning benchmarks.
Waseda University, Adelaide University · Sept 2026
- Chinese Techcs.AI
AI agents get nearly ten times more training tasks from scientific code
Poseidon generates nearly ten times more training environments from scientific code repositories than prior methods and improves agent performance.
The Hong Kong University of Science and Technology (Guangzhou), Hunyuan Team, Tencent · Oct 2026
- Big Techcs.CL
Language models improve using self-generated questions and answers, without human examples
Large language models self-improve using only self-generated questions and privileged reasoning, without human-curated data.
Google, Ulm University · Oct 2026
- Research Labcs.LG
Self-teaching language models improve when the teacher gets feedback too
Teacher-Evolution Self-Distillation (TESD) enhances on-policy self-distillation by improving self-teacher supervision, outperforming existing methods.
Chinese Academy of Sciences, University of Chinese Academy of Sciences · Oct 2026
- Big Techcs.CL
Language models learn to propose improvements for other language models via RL with a learned world model
LM2LM uses reinforcement learning to train LLMs to generate effective proposals for improving other language models.
The Pennsylvania State University, UC Santa Cruz · Oct 2026
- Top Universitycs.AI
Tailoring agent harnesses per task beats one-size-fits-all optimization
Turbo Harness adapts global harnesses per instance using prior optimization artifacts, outperforming baselines across seven benchmarks.
Rutgers University, Red Hat AI Innovation · Oct 2026
- Big Techcs.AI
AI progress may continue past AGI toward superintelligence, not stop
Post-AGI progress to artificial superintelligence could occur through scaling, paradigm shifts, recursive improvement, and multi-agent collectives, with uncertain bottlenecks.
Google DeepMind, University of Waterloo · June 2026
- Chinese Techcs.AI
Language model and skill set co-evolve through response feedback
SkillDuet coordinates model parameter updates and skill retention using response feedback, achieving 98.57% on ALFWorld and 92.97% on WebShop.
Alibaba · Nov 2026
- Top Universitycs.LG
Recursive self-distillation on benign outputs can progressively misalign safe models
Recursive self-distillation with benign cues causes progressive misalignment in safe language models.
University of California, Berkeley, ETH Zurich · Oct 2026
- Top Universitycs.LG
Agent harness optimization mostly improves execution reliability, not task reach
Gains from agent harness optimization are largely elicitation, not expansion; pass@1 advantages over stronger systems reverse at larger attempt budgets and vanish on hard tasks.
FuriosaAI, Yonsei University · Oct 2026
- Big Techcs.LG
Prompt evolution lets frozen LLMs learn new forecasting regimes without retraining
Evolving a short guidance prompt lets a frozen LLM outperform time-series foundation models on held-out multimodal forecasting without retraining.
Amazon Web Services, University of California · Oct 2026
- Chinese Techcs.AI
Diagnosis-first framework boosts solver performance across numerical domains
ADSD links numerical diagnosis to reusable solver skill discovery, improving solver accuracy, robustness, and efficiency across four numerical domains.
University of California, Berkeley, Alibaba Group · Oct 2026
- Chinese Techcs.AI
Structuring noisy file environments improves agents’ performance on complex tasks
Env-Rethink prepares and evolves agent environments using collection maps and event logs, lifting mean rubric pass rate by 15.1 points across nine models.
Shanghai Jiao Tong University, Theseus Labs · Sept 2026
- Top Universitycs.SE
Agents write verifiers that wrongly reject valid solutions more often than they accept them
LLM agents generate verifiers that reject invalid solutions more reliably than they accept valid ones.
University of Pennsylvania, University of Illinois at Urbana-Champaign · Oct 2026
Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.