Subscribe
Share
Share
Embed
Claude Opus 4.8: Dynamic Workflows, Hundreds of Agents, and the Benchmarks That Matter
The script explains Claude Opus 4.8’s release, highlighting “dynamic workflows” that let Claude spin up hundreds of coordinated agents in one session to plan, build, review, and self-check work, with agents able to run for days. It describes enabling this via Claude Code’s Ultra Code mode (and by asking Claude to create a dynamic workflow), and promotes an “agent operating system” inside the AI Profit Bot/AI Profit Boardroom with a zip file, tutorial, 30-day roadmap, prompts, weekly coaching calls, and a community. The video reviews benchmarks: SWE-bench Verified 88.6% and SWE-bench Pro 69.2% (ahead of GPT-5.5 and Gemini 3.1 Pro), Terminal Bench 74.6% (behind GPT-5.5), Frontier SWE rank #1, office-work GTP-Val where 4.8 beats GPT-5.5 in most matchups, improved Zapier workflow score, and a major long-context memory jump. It also notes reduced overclaiming, lower “sneaky” behavior, effort controls, faster/cheaper fast mode, and that Anthropic’s teased Mythos model may arrive soon.
00:00 Opus 4.8 Biggest Upgrade
00:52 Dynamic Workflows Explained
01:25 Ultra Code One Click Setup
01:51 Parallel Agent Speed Gains
02:33 Bun Rebuild Case Study
03:10 Agents Running For Days
03:33 Agent OS And Offer
04:26 Coding Benchmarks Breakdown
06:08 Office Legal Workflow Tests
07:17 Long Context Memory Leap
08:22 Honesty And Safety Improvements
09:15 Mythos Preview And Timeline
10:02 Effort Controls Pricing Speed
10:34 Wrap Up And Next Steps
The script explains Claude Opus 4.8’s release, highlighting “dynamic workflows” that let Claude spin up hundreds of coordinated agents in one session to plan, build, review, and self-check work, with agents able to run for days. It describes enabling this via Claude Code’s Ultra Code mode (and by asking Claude to create a dynamic workflow), and promotes an “agent operating system” inside the AI Profit Bot/AI Profit Boardroom with a zip file, tutorial, 30-day roadmap, prompts, weekly coaching calls, and a community. The video reviews benchmarks: SWE-bench Verified 88.6% and SWE-bench Pro 69.2% (ahead of GPT-5.5 and Gemini 3.1 Pro), Terminal Bench 74.6% (behind GPT-5.5), Frontier SWE rank #1, office-work GTP-Val where 4.8 beats GPT-5.5 in most matchups, improved Zapier workflow score, and a major long-context memory jump. It also notes reduced overclaiming, lower “sneaky” behavior, effort controls, faster/cheaper fast mode, and that Anthropic’s teased Mythos model may arrive soon.
00:00 Opus 4.8 Biggest Upgrade
00:52 Dynamic Workflows Explained
01:25 Ultra Code One Click Setup
01:51 Parallel Agent Speed Gains
02:33 Bun Rebuild Case Study
03:10 Agents Running For Days
03:33 Agent OS And Offer
04:26 Coding Benchmarks Breakdown
06:08 Office Legal Workflow Tests
07:17 Long Context Memory Leap
08:22 Honesty And Safety Improvements
09:15 Mythos Preview And Timeline
10:02 Effort Controls Pricing Speed
10:34 Wrap Up And Next Steps
Creators and Guests
What is AI News Today | Julian Goldie Podcast?
Latest Podcast