long horizon
30 articles about long horizon in AI news
Chamath: Long-Horizon AI Tasks Are a Joke, Trough Ahead
Chamath Palihapitiya at Stanford AI Club says long-horizon AI tasks don't work, warns of trough of disillusionment, and proposes symbolic space approach.
Microsoft Paper Probes Long-Horizon Agent Generalization Gap
Microsoft Research paper on long-horizon agent generalization identifies failure modes and proposes improvements for extended tasks.
Researchers Achieve Ultra-Long-Horizon Agentic Science with Cohesive AI Agents
A research team has developed AI agents capable of executing and maintaining coherent, long-horizon scientific research workflows. This addresses a core challenge in creating autonomous systems for complex discovery.
HORIZON Benchmark Diagnoses Long-Horizon Failures in GPT-5 and Claude Agents
A new benchmark called HORIZON systematically analyzes where and why LLM agents like GPT-5 and Claude fail on long-horizon tasks. The study collected over 3100 agent trajectories and provides a scalable method for failure attribution, offering practical guidance for building more reliable agents.
OpenResearcher Paper Released: Method for Synthesizing Long-Horizon Research Trajectories for AI
The OpenResearcher paper has been released, exploring methods to synthesize long-horizon research trajectories for deep learning. This work aims to provide structured guidance for navigating complex, multi-step AI research problems.
AI Agents Get a Memory Upgrade: New Research Tackles Long-Horizon Task Challenges
Researchers have developed new methods to scale AI agent memory for complex, long-horizon tasks. The breakthrough addresses one of the biggest limitations in current agent systems—their inability to retain and utilize information over extended sequences of actions.
EnterpriseArena Benchmark Reveals LLM Agents Fail at Long-Horizon CFO-Style Resource Allocation
Researchers introduced EnterpriseArena, a 132-month enterprise simulator, to test LLM agents on CFO-style resource allocation. Only 16% of runs survived the full horizon, revealing a distinct capability gap for current models.
Superforecasters Predicted 3-4h AI Task Horizons by Year-End; Claude Hit It in May
Superforecasters predicted 3-4h METR 80% task horizons by year-end 2026. Claude Mythos hit that in late May, compressing the timeline by seven months.
Claude Mythos Preview Doubles METR Time Horizon at 80% Success
Claude Mythos Preview snapshot achieves 2x METR time horizon over next best model at 80% success rate, per Anthropic. Absolute numbers undisclosed.
Gemini 3.1 Pro Leads METR Time Horizon, Handles 90-Minute Software Tasks
Google's Gemini 3.1 Pro is the new leader on METR's time horizon benchmark, successfully handling software tasks that take humans an average of 1 hour and 30 minutes to complete, with an average score of 77%. This marks a significant shift as Google takes the top spot from OpenAI and Anthropic on a key benchmark measuring autonomous agent capability.
Sam Altman Envisions AI That Thinks for Days: The Dawn of Super-Long-Term Reasoning
OpenAI CEO Sam Altman predicts future AI models will perform "super long-term reasoning," spending days or weeks analyzing complex, high-stakes problems. This represents a fundamental shift from today's rapid-response systems toward deliberate, extended cognitive processes.
AI's Time Horizon Expands: Claude and GPT Push Multi-Hour Task Capabilities
New analysis reveals Claude Opus 4.6 and GPT 5.3 Codex can handle complex tasks requiring hours of human effort. The METR benchmark shows AI systems approaching 3-4 hour time horizons at 50% success rates, signaling major progress in sustained reasoning.
PlanBench-XL: GPT-5.4 Scores 11.36% on Hard Tool-Use Tasks
PlanBench-XL shows GPT-5.4 drops from 51.90% to 11.36% accuracy on long-horizon tool-use tasks with 1,665 tools, revealing a fundamental planning weakness.
Sleep Phase Cuts Transformer Costs by Consolidating Memory
Paper proposes sleep phase to consolidate context into fixed-size memory, reducing inference cost while improving long-horizon task performance on GSM-Infinite.
ML-Master 2.0 Hits 56.44% on MLE-Bench in 24-Hour Agentic Science Run
Researchers from Shanghai Jiao Tong University demonstrated ML-Master 2.0, an autonomous research agent that operated continuously for 24 hours on the MLE-Bench, achieving a 56.44% medal rate. The breakthrough centers on Hierarchical Cognitive Caching for state management, not reasoning, enabling long-horizon scientific workflows.
NVIDIA Lyra 2.0 Launches on Hugging Face for Persistent 3D World Generation
NVIDIA has released Lyra 2.0 on Hugging Face, a framework designed to generate persistent, explorable 3D worlds at scale. It specifically addresses the core technical challenges of spatial forgetting and temporal drifting in long-horizon video generation.
Inside Claude Code’s Leaked Source: A 512,000-Line Blueprint for AI Agent Engineering
A misconfigured npm publish exposed ~512,000 lines of Claude Code's TypeScript source, detailing a production-ready AI agent system with background operation, long-horizon planning, and multi-agent orchestration. This leak provides an unprecedented look at how a leading AI company engineers complex agentic systems at scale.
MiRA Framework Boosts Gemma3-12B to 43% Success Rate on WebArena-Lite, Surpassing GPT-4 and WebRL
Researchers propose MiRA, a milestone-based RL framework that improves long-horizon planning in LLM agents. It boosts Gemma3-12B's web navigation success from 6.4% to 43%, outperforming GPT-4-Turbo (17.6%) and the previous SOTA WebRL (38.4%).
ServiceNow Research Launches EnterpriseOps-Gym: A 512-Tool Benchmark for Testing Agentic Planning in Enterprise Environments
ServiceNow Research and Mila have released EnterpriseOps-Gym, a high-fidelity benchmark with 164 database tables and 512 tools across eight domains to evaluate LLM agents on long-horizon enterprise workflows.
TimeSqueeze: A New Method for Dynamic Patching in Time Series Forecasting
Researchers introduce TimeSqueeze, a dynamic patching mechanism for Transformer-based time series models. It adaptively segments sequences based on signal complexity, achieving up to 20x faster convergence and 8x higher data efficiency. This addresses a core trade-off between accuracy and computational cost in long-horizon forecasting.
Accenture's Memex(RL) Revolutionizes AI Agent Memory for Complex Tasks
Accenture researchers have developed Memex(RL), a breakthrough system that gives AI agents structured, searchable memory for long-horizon tasks. This solves the critical problem of agents losing track of past experiences during complex operations like deep research and multi-step planning.
Neural Paging: The Memory Management Breakthrough for Next-Gen AI Agents
Researchers propose Neural Paging, a hierarchical architecture that decouples symbolic reasoning from information management in AI agents. This approach dramatically reduces computational complexity for long-horizon reasoning tasks, moving from quadratic to linear scaling with context window size.
PseudoAct: How Pseudocode Planning Could Revolutionize AI Agent Decision-Making
Researchers have developed PseudoAct, a new framework that enables AI agents to plan complex tasks using pseudocode before execution. This approach addresses critical limitations in current reactive systems, reducing redundant actions and improving efficiency in long-horizon tasks by up to 20.93%.
The Hidden Culprit in AI Agent Failure: New Research Reveals Surprising Pattern
A new study challenges conventional wisdom about why AI agents fail in complex tasks, finding that most failures stem from forgetting earlier instructions rather than insufficient knowledge. This discovery has significant implications for developing more reliable long-horizon AI systems.
Intology's Locus beats human-tuned Qwen3-1.7B in auto post-training
Intology's Locus beat human-tuned Qwen3-1.7B (51.6% vs 49.4%) on PostTrainBench by scaling compute 64x, showing AI research agents need longer timescales.
Federated MCP Servers: How to Scale Claude Code from Monolith to
Federated MCP networks turn Claude Code from a monolith into a microservices mesh. Use a Supervisor agent + specialized MCP servers (Stdio/SSE transports) to cut integration complexity from O(N²) to O(N) and scale horizontally.
OpenAI Launches GPT-Rosalind for Drug Discovery, GPT-5.4-Cyber for Security
OpenAI launched GPT-Rosalind, a life sciences model performing above the 95th percentile of human experts on novel biological data, and GPT-5.4-Cyber, a cybersecurity variant. These releases, alongside a major Agents SDK update, signal a pivot from general AI to specialized, high-stakes enterprise domains.
Goal-Aligned Recommendation Systems: Lessons from Return-Aligned Decision Transformer
The article discusses Return-Aligned Decision Transformer (RADT), a method that aligns recommender systems with long-term business returns. It addresses the common problem where models ignore target signals, offering a framework for transaction-driven recommendations.
Meta Ties Executive Bonuses to $9 Trillion Valuation Target by 2031, Aligning with AI Ambitions
Meta has informed senior executives their full long-term compensation is contingent on the company reaching a $9 trillion market valuation by 2031. This aggressive target underscores the financial scale of its AI and metaverse bets.
New RL-Guided Planning Framework Boosts Warehouse Robot Throughput
Researchers propose RL-RH-PP, a hybrid AI framework combining reinforcement learning with classical search for lifelong multi-agent path finding. It dynamically assigns robot priorities to reduce congestion, achieving higher throughput in simulations and generalizing across layouts.