Claude Opus 4.8: "Ultra Code" Unleashed for Complex Tasks

Summary

Anthropic's latest Claude Opus 4.8 model introduces "Ultra Code," a powerful new feature allowing it to tackle immensely complex, long-term tasks by orchestrating hundreds of parallel agents. This upgrade significantly enhances agent reliability, enabling them to run for extended periods and with greater honesty. Early tests show Opus 4.8 outperforming previous versions and competitors on various coding and general intelligence benchmarks, including a significant leap in its ability to identify and report code flaws. One compelling example is its successful porting of the "bun" project to Rust in just eleven days, handling roughly seven hundred fifty thousand lines of code with a remarkably high pass rate on existing tests. While pricing for the API remains consistent, a "fast mode" has become three times cheaper and twice as fast. Anthropic also teases upcoming, lower-cost models with similar capabilities and a future "Mythos" model designed to surpass Opus in intelligence, hinting at a new class of AI.