https://frontieraianalysis.com/articles/apertus15-open-stages-multimodal-audit/Frontier AI Analysisen2026-10-21T04:14:06-04:00Apertus 1.5 exposes the stages behind its multimodal expansionhttps://frontieraianalysis.com/articles/exa3-highlight-selection-evidence-budget/Frontier AI Analysisen2026-10-21T04:14:06-04:00Exa 3.0 shifts the search budget toward selected evidencehttps://frontieraianalysis.com/articles/fireworks-multiregion-placement-capacity/Frontier AI Analysisen2026-10-21T04:14:06-04:00Fireworks separates global capacity placement from the inference request pathhttps://frontieraianalysis.com/articles/giskard-stereotales-generation-versus-recognition/Frontier AI Analysisen2026-10-21T04:14:06-04:00Giskard’s StereoTales tests what models generate, not only what they recognizehttps://frontieraianalysis.com/articles/huggingface-agent-curriculum-training-signal/Frontier AI Analysisen2026-10-21T04:14:06-04:00Hugging Face’s agent curriculum separates imitation from learning through actionhttps://frontieraianalysis.com/articles/lighton-slideon-sparse-retrieval-operating-budget/Frontier AI Analysisen2026-10-21T04:14:06-04:00LightOn’s SlideOn makes retrieval sparsity an adjustable operating choicehttps://frontieraianalysis.com/articles/linkup-aleph-alpha-search-data-boundary/Frontier AI Analysisen2026-10-21T04:14:06-04:00Linkup’s Aleph Alpha case study puts search inside the data boundaryhttps://frontieraianalysis.com/articles/modal-candidates-mirror-selection-bias/Frontier AI Analysisen2026-10-21T04:14:06-04:00Modal’s Endpoint Candidates make inference changes testable beside productionhttps://frontieraianalysis.com/articles/patronus-speedrunbench-budget-route-discovery/Frontier AI Analysisen2026-10-21T04:14:06-04:00Patronus’ SpeedrunBench asks whether an agent can improve a successful routehttps://frontieraianalysis.com/articles/prime-rl010-policy-mismatch-training/Frontier AI Analysisen2026-10-21T04:14:06-04:00Prime Intellect targets the mismatch between agent rollouts and training