reasoning models
6 published articles
Test-Time Scaling
CoBa routing matches best-of-16 voting with 58.9% fewer tokens
CoBa, a compute-balanced routing policy from a new arXiv paper, matches best-of-16 majority voting within 0.01 points while cutting parameter-weighted tokens by 58.9%. On 3,129 evaluations across MATH-500, AIME, and AMC, it also beats single-sample decoding outright, though a small best-of-16 edge remains when budget is no object.
2026-08-19
AI Research
Longer chain-of-thought hits a wall. ThinkRetrieve injects the fix mid-reasoning
Sequential test-time scaling often hits diminishing or even negative returns, a new preprint argues. ThinkRetrieve retrieves solved examples mid-reasoning and injects them into the trace, reporting relative gains up to 60% on AIME 2025 across five small reasoning models.
2026-08-18
AI Research
The monitor that goes silent when AI reasoning fails
Novelis Research shows token log-probability fails as a decoder monitor for quantized reasoning models, being blind to confident loops. They introduce a calibrated e-CUSUM controller that combines uncertainty and repetition signals, achieving selectivity on GSM8K with DeepSeek-R1-Distill-Qwen-1.5B.
2026-08-03
Artificial Intelligence
The hardest lesson for AI reasoning engines: when to shut up
MIT researchers propose OS-Pruner, a plug-in that dynamically stops chain-of-thought reasoning when further computation isn't worth the token cost. Tests show 20-60% length reduction with minimal accuracy sacrifice.
2026-07-29
Benchmark Deep Dive
ARC-AGI-2: The Benchmark That Measures Fluid Intelligence in AI Systems
ARC-AGI-2 tests AI systems on fluid intelligence through visual grid puzzles that can't be solved by memorization. Top frontier models now score 75-85%, but the grand prize of $700,000 remains unclaimed. Here's a deep dive into the benchmark's design, scoring, and current leaderboard.
2026-07-01
AI Safety Research
AI models can't stop thinking out loud. That's both good news and a nightmare for safety.
Claude Sonnet 4.5 can control its chain-of-thought only 2.7% of the time, versus 61.9% for final outputs. The gap raises open questions about the robustness of CoT monitoring as a safety mechanism, and nobody knows why it exists.
2026-03-09