Its biggest change is that it's more honest.
Source: X platform
Anthropic today released Claude Opus 4.8, priced the same as its predecessor, 4.7.
Its biggest improvement is greater honesty: it’s more willing to admit uncertainty, less likely to fabricate answers just to provide one, and better at accurately assessing its own progress. When handling long-running agent tasks, it behaves more like a reliable engineer—you don’t need to constantly monitor it.
Also launching today is Fast Mode, which offers approximately 2.5x faster speed with the same model and a price three times lower than before. Enable it in Claude Code using /fast; API users must contact their account manager to request access or join the waiting list.
The highlight is dynamic workflows.
Along with Opus 4.8, Claude Code is introducing a new feature called dynamic workflows, currently in research preview.
You give it a large task, and it automatically breaks it down, deploying dozens to hundreds of parallel subagents to handle it. Once completed, another set of agents verifies the results—even assigning specific agents to scrutinize and challenge the outputs—iterating repeatedly until the results converge, ultimately delivering you a consolidated, refined answer. The entire process can run for hours or even days, and if interrupted, it can seamlessly resume from where it left off.
Suitable tasks: Debugging the entire codebase, conducting security audits, optimizing performance, and handling the most common large-scale migrations, including framework upgrades, API replacements, and cross-language porting—projects involving thousands of files at once.
Anthropic promoted the rewrite of Bun in Rust as a selling point. Bun is a fast JavaScript runtime; its creator, Jarred Sumner, migrated the entire project from Zig to Rust using dynamic workflows. The official statement claims approximately 750,000 lines of Rust code were written, passing 99.8% of the original tests, and the initial commit to merge took just 11 days.
Cost: It consumes a lot of tokens.
Anthropic has unusually issued a proactive warning: dynamic workflows consume significantly more tokens than standard Claude Code sessions; it is recommended to start with small tasks. When triggered for the first time, Claude Code will display what it intends to execute for your confirmation, and enterprise administrators can also disable it directly.
Max, Team plans, and API users have this feature enabled by default; for Enterprise plans, it is disabled by default and must be manually enabled by an administrator. To enable it, simply ask Claude to “create a workflow” or toggle the ultracode switch.