Opus 4.8 sets a new state of the art on ARC-AGI-3
Opus 4.8 is new SOTA on ARC-AGI-3!
It was fun to help contribute to the open source benchmark harness for running this eval.
Github repo: github.com/arcprize/arc-agi-3-benchmarking
- Models and research
Keep reading
Newer: Anthropic engineers now ship 8x as much code per quarterModels and research
Anthropic is already profitable on inference unit economics
Anthropic is already profitable on inference unit economics. anthropic.com/news/confidential-draft-s1-sec
Coding agentsAI agents
Claude Code dynamic workflows move abstraction up a level
Claude releases workflows. This moves the abstraction layer one level up again. Rather than having to craft a multi-agent harness for your codebase, this allows Claude to intelligently come up with the harness. The workflow is modeled as a DAG with agents as nodes. So its a generic computation construct. Elegant! claude.com/blog/introducing-dynamic-workflows-in-cl…
AI agents
Congrats to Fang Cheng and Caeli on the launch
Excited for this launch Fang Cheng. These types of tedious yet simple rules based decision making is what AI agents are perfect for.
Customers and partners
Congrats to Cargo Robotics on signing a top parcel carrier
Congrats Daniel and Cargo team! Always exciting to see a StartX portco and a Plank customer be successful! LFG
Models and researchAI agents
Anthropic's Mythos security concerns are not marketing spin
Great to meet the Anthropic team at the Claude Code conference...Specially enjoyed the conversations with the engineers working on managed agents and long-running agents . Also they are genuinely scared about the security risks of releasing Mythos, it's not just marketing spin or compute constraints. claude.com/code-with-claude/san-francisco-extended
Coding agents
Attending Anthropic's Code with Claude in San Francisco
Excited to join the Anthropic developer conference in San Francisco! You can watch the livestream if you cannot be there in person. claude.com/code-with-claude/san-francisco
Enterprise AIAI agents
Anthropic finance agents show vertical SaaS is just agents
Anthropic releases agents for various financial services and insurance use cases. Most of vertical SaaS, at its essence, is these agents. Anthropic is making Claude the interface for knowledge workers and different use cases to just be agents created by domain experts. The agents are connected with underlying systems of record. What is missing in this architecture? anthropic.com/news/finance-agents
Enterprise AIForward deployed engineering
Anthropic launches an enterprise AI services firm
Interesting times ahead! There’s a $1T+ opportunity here to fundamentally transform the operational playbook of running a business sequoiacap.com/article/services-the-new-software