LSA predicts relevant context sections in advance and retains only these in GPU memory, compressing the KV-cache by over 86 percent without sacrificing accuracy.
Claude Fable 5 demonstrates significant performance improvements over predecessor models, while Anthropic simultaneously tightens access controls that set a regulatory precedent for the industry.
LCLMs compress KV-caches through encoder-decoder architecture up to 1:16 more efficiently than previous methods while reducing peak memory consumption and processing time.
Encoder-decoder compressors with adaptive expansion improve KV-cache compression methods in speed and memory efficiency without significant quality loss.
Project Headroom filters redundant data from API requests to reduce token costs – users report estimated savings of $700,000 and 200 billion tokens since January 2026.
Reasoning Arena replaces uninformative rewards with head-to-head comparisons of solution attempts and reduces required compute time by 27 to 41 percent.
Optical reasoning uses images as the primary reasoning medium, saving an average of 28.57 percent tokens on language tasks and 16 percent on multimodal tasks.
Fable 5 sets new benchmarks in software engineering and knowledge work through extended autonomous runtimes, while Mythos 5 offers cybersecurity capabilities without security restrictions.
Anthropic offers Fable 5, a Mythos variant with safety filters for public use, while Project Glasswing participants gain access to less restricted Claude Mythos 5, accompanied by new federal rules controlling frontier AI models.
Anthropic publicly releases the more powerful Claude variant Fable 5, but automatically routes potentially dangerous cybersecurity requests to a weaker model.
