Anthropic Releases Opus 4.8 With New 'Dynamic Workflow' Tool
- Reference: 0183438094
- News link: https://slashdot.org/story/26/05/28/2013200/anthropic-releases-opus-48-with-new-dynamic-workflow-tool
- Source link:
> Opus 4.8 comes with the expected best-in-class benchmark results, but there's also particular attention to how the model manages bad or uncertain data. In the launch post, Anthropic's early testers found that the new model is "more likely to flag uncertainties about its work and less likely to make unsupported claims." Echoing this point, a testimonial from Bridgewater associates said the biggest difference in the upgrade was "Opus 4.8's tendency to proactively flag issues with the inputs and outputs of an analysis, something other models routinely missed and left to the users to catch."
>
> Together with the new model, Anthropic launched a feature called Dynamic Workflows, which will be available in research preview. The system is designed to help larger models like Opus manage complex tasks across hundreds of parallel subagents. "Claude Code alongside Opus 4.8 can now carry out codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge, with the existing test suite as its bar," the post explains.
As for [3]Mythos , Anthropic's most advanced model, the company hinted it could be made publicly available in the not too distant future. "We're making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks," the company wrote.
[1] https://www.anthropic.com/news/claude-opus-4-8
[2] https://techcrunch.com/2026/05/28/anthropic-releases-opus-4-8-with-new-dynamic-workflow-tool/
[3] https://it.slashdot.org/story/26/04/07/2115208/anthropic-unveils-claude-mythos-powerful-ai-with-major-cyber-implications
Maybe now we can finally get rid of COBOL? (Score:2)
I wonder how *that* would go!
There's some weird stuff in legacy code. Like legacy Unisys COBOL, which has a 48-bit word. What could possibly go wrong?
Re: (Score:2)
Anthropic's products are more likely to extend the life of COBOL than shorten it.
Re: (Score:2)
Good point!
Re: (Score:2)
There are 80-bit extended double types supported in the most modern Intel CPU's so what's your point?
Properly written software doesn't cause problems.
That is why COBOL is still around and most of the internet is supported by software written in even more exotic languages like Erlang.
Re: (Score:2)
You seem to be young, imagining that software in those days was "properly" written, especially compared to today's standards. It was common, for example, to allocate an array of 48-bit words, and then place text in that buffer as 8-bit bytes, completely ignoring the word boundaries. Another fun technique was storing a string of bytes in one of these buffers, and then executing it as machine code.
It was the wild, wild west.
Mass code migrations are full of risk (Score:4, Interesting)
A big reason we write code in small chunks, is so we can review the changes and be sure it works as expected. Bulk-testing a large project automatically converted by *any* tool, AI or otherwise, is bound to have a million unexpected issues that are impossible to test, because there are too many scenarios to cover.
Wholesale migrations is never a good idea, if risk matters.
Ok I'm now interested (Score:3)
If it can go "I don't know" rather than spew bollocks as fact it may actually be the turning point for AI that would get me to use it. At least for stuff I don't already know about already. Will have to try with things the previous one claimed to know despite being provably wrong.
Re: oh geeeeez.... (Score:1)
If anyone is paying attention, the vultures are circling around Anthropic. Apollo, Blackstone and Broadcom are on the case: [1]https://www.reuters.com/busine... [reuters.com]
[1] https://www.reuters.com/business/apollo-blackstone-work-36-billion-debt-deal-anthropic-bloomberg-news-reports-2026-05-28/
Re: (Score:2)
my guy.
It's not 'offended' when some naive biologist uses this workflow engine and has a protein name interpreted as a date and then expects me to be able to fix it because they associate me with the computer.
Re: (Score:2)
Lol as if any codebase that needs this actually has tests.