Parallel agent teams scale test-time compute better
Parallel teams of ten outperform best-of-N and large single teams on long-horizon tasks with a 27B model
Stories per day · 30 days
2 RSS
Relatedagents1ai-agents1anthropic1automated-verification1benchmarks1formal-verification1inference-cost1lean1llm-agents1math1openai1reasoning-effort1theorem-proving1
19 stories, newest first · one card per story
Stories tagged #test-time-compute
Higher LLM reasoning effort can cost more without better results
On easy tasks every effort level solves, the highest setting costs up to 42 times the lowest, and escalating after failure is often cheaper.
Simple Lean 4 harness h1 matches specialized prover workflows
With matched models, a single attempt plus verifier feedback solved eight open FormalConjectures problems and led three formal benchmarks.
Essay says verifiers let AI search hard math problems depth-first
The author contrasts months-long human search with models that can test weak hints in minutes if a checker rejects false proofs.
1
PreSTO: Predictive Subtree Prefetching for Fast LLM Power Sampling
PreSTO speeds PowerMH LLM sampling by prefetching proposal subtrees without changing the transition kernel.
Scaling to Tens of Thousands of Test-Time Iterations with Loop-Native Attention Residuals
InfiLoop adds loop-native residuals so looped Transformers keep improving past 20,000 test-time steps.
RefCon: Iterative Refinement and Contrastive Memory Extraction for Context-Evolving Agent
RefCon extracts agent memories via self-refinement and contrast, beating unlabeled baselines on AppWorld and BFCL-V3.
Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents
Mid-Harness samples and verifies terminal-agent actions before execution, lifting Pass@1 on TerminalBench-Lite.
Scheduling Recursive Reasoning in Looped Transformers
TAPS adapts recurrent step size in looped transformers, raising accuracy and cutting loops without retraining.
Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI
Frozen Builder models learn reusable Meta-Skills that improve agent harnesses by 8.95 points on Harness-Bench and NewtonBench.
Why Deterministic PRM Guidance Underperforms in Discrete Diffusion Reasoning
Deterministic PRM guidance for discrete diffusion LLMs loses to simpler ORM reranking at matched compute.
Beyond Repeated Sampling: Learning Search Policies for LLM Reasoning
An RL-trained concept generator lifts a frozen LLM's pass@k on hard math versus repeated sampling.
When Agents Slow Down: Understanding LLM Agents' Test-Time Strategies via Elo-per-token Analysis
Elo-per-token analysis shows LLM agents' marginal gains drop below independent sampling at scale; parallel sessions beat one long session.
[AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded
OpenAI-linked accounts claim roughly 10,000 AI agents produced a Navier-Stokes singularity result in 88 hours, pending mathematical verification.
4
Studying Without a Syllabus: Task-Agnostic Environment Preprocessing
Paper formalizes task-agnostic environment preprocessing, where agents study unfamiliar environments under a budget to build reusable artifacts for a frozen solver.
An Open Recipe for IMO Gold: Training Nemotron for Olympiad Mathematics
Open post-training pipeline turns Nemotron 3 Ultra checkpoints into an IMO 2026 gold-medal system, scoring 30/42 without formal provers or external tools.
A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM
A*-Thought-V2 compresses redundant chain-of-thought steps into latent tokens guided by hidden-state geometry, improving accuracy up to 2.6% while halving response length.
Rethinking Indirect Prompt Injection as a Test-Time Search Problem
Researchers frame indirect prompt injection as test-time search, showing added attacker compute improves vulnerability discovery and exploitation against tool-using agents.
What Else Needs Fixing? Exploring Cost-Effective Test-Time Compute for Revision Propagation in Artifacts Generated Through Conversation
A new benchmark shows LLMs reach 68.3-93% accuracy propagating local revisions across conversationally generated artifacts, with parallel-sample selection most cost-effective.