Infrathon tests language agents on long-horizon AI infrastructure tasks
Across 150 runs and over 10,000 GPU-hours, 135 still improved in the second half of a 24-hour budget.
Stories per day · 30 days
2 RSS
Relatedreinforcement-learning5benchmarks2sample-efficiency2agent-interfaces1agent-memory1agent-skills1ai-infrastructure1alfworld1backtesting1benchmark1browsecomp1cli1context-compression1cope1credit-assignment1dart1diffusion-models1evaluation1
15 stories, newest first · one card per story
Stories tagged #language-agents
FINCOSMOS evaluates finance agents without fixed tasks, data, or timelines
In 150 sessions no run passed; agents surfaced only 54-64% of untold details, and experts blamed them for 88.33% of failures.
SUMONE summarizer compresses agent histories by 78% on DeepSearchQA
SUMONE summarizes trajectories via masked reconstruction and GRPO reward for task-agnostic plug-and-play compression.
EvoGuard improves agent skill evolution by 11% on ALFWorld
EvoGuard uses pairwise rejection sampling on private test sets to select non-degrading skills at low latency.
Metacognitive controller trims diffusion language-agent inference cost
Across five benchmarks, the plug-in cuts inference FLOPs 20.6% on average versus the strongest efficient baseline while keeping competitive accuracy.
EnvSched schedules agent RL rollouts by environment and task
On Qwen3-8B it scores 48.81 after 40 steps, above random selection's best score after 120 steps.
FOCUS uses selective branching to train long-horizon agent critics
At Qwen3.5-9B, held-out accuracy rises 7.2 points on BrowseComp-Plus and 4.1 on DeonticBench versus the strongest baseline.
Skill-infinity makes agent skill optimization self-improving
On 83 SkillsBench tasks the authors report 60.7% pass rate, above curated skills and SkillEvolver.
Matched MCP and CLI interfaces change language-agent behavior, study finds
DART improved scores by up to 9.0 points and cut MCP cost by up to 27.1k tokens per task.
Warm-Start UCB personalizes memory gating for language agents
The policy learns which facts to keep from delayed retrieval feedback and beat baselines across memory budgets.
CoPE co-trains action forecasting with language-agent policies
On ScienceWorld the authors report up to 1.9 times and 2.0 times the sample efficiency of ECHO and GRPO.
Co-Evolving Acting and Forecasting for Language Agents
CoPE trains language agents to act and forecast future actions, improving benchmark scores and sample efficiency.
Towards Communication-Efficient Social Intelligence in Language Agents
TACT trains language agents to reach social goals with fewer tokens via teacher-assisted strategy and expression revisions.
Cognitive Extensions for Dual-Process Language Agents: Memory and Self-Reflection in Interactive Environments
Researchers extend the SwiftSage dual-process agent with adaptive memory and self-reflection modules, improving scores in interactive environments.
ResidualAuth: What Authorization State Must Language Agents Preserve under Revocable Delegation?
Formalizes residual authorization state language agents must preserve under revocable delegation; token-budget summaries mostly fail while hard gates stop unauthorized effects.