Researchers submitted CliffCompaction, an autocompaction technique for long-horizon coding agents, to arXiv on 22 Sep 2026, reporting cost reductions of up to 50% under a bounded context while maintai…
Published semantic-layer benchmarks from Snowflake, dbt, Cube, and Denodo conflate two distinct claims — richer business context versus deterministic SQL compilation — so their reported accuracy lifts…
Neon's benchmark of 42 AI models on a synthetic support-ticket workload found GPT-5 Nano was the cheapest at $0.947 per 100 tickets, while GPT-5.5 Pro was the most expensive at $8.34, and GPT-5.3 Code…
DogLM, a new benchmark created by Mike Ushakov, tests whether large language models (LLMs) let players pet a dog in AI-generated video games, inspired by the 'Can You Pet the Dog?' rule. In tests of 1…