gekro
GitHub LinkedIn
News

AI News

Gigatoken tokenizer hits 24.5 GB/s; Cursor Router cuts inference costs 30–50%

A new Rust tokenizer achieves near-gigabyte-per-second throughput; Cursor's request router cuts frontier-model costs through dynamic model selection.

1 min read 3 sources

Gigatoken, an MIT-licensed Rust BPE tokenizer, encodes text at 24.53 GB/s on a 144-core AMD EPYC processor, achieving 989x speedup over HuggingFace Tokenizers and 681x over tiktoken (MarkTechPost). The gains stem from hand-optimized multithreaded Rust implementation, offering material impact for token-rate-limited inference pipelines and batch processing workloads.

Cursor has made Cursor Router generally available, a request-level classifier that routes queries to the most suitable model based on complexity and task domain, delivering frontier-quality output at 60% savings in A/B tests and 30–50% cost reduction in production deployment (MarkTechPost). In parallel, Anthropic released Claude Security as a beta plugin for Claude Code - a multi-agent vulnerability scanner operating from the terminal that flags issues and auto-generates patches for review (MarkTechPost).

Compiled automatically from the linked sources and published without manual editing - a neutral summary of third-party reporting, for information only. Every claim links to its origin. Not original reporting.