Opus 4.7 retook the top spot for agentic coding among publicly available models, scoring 64.3% on SWE-bench Pro and SWE-bench Verified. It arrives two months after Opus 4.6, continuing a steady upgrade cadence, and reportedly pulls ahead of GPT-5.4 and Gemini 3.1 Pro on key benchmarks. Pricing stays at $5 per million input tokens and $25 per million output tokens, and the model is live across Claude products, the API, Bedrock, Vertex AI, and Microsoft Foundry.
Why It Matters: There's a real shift happening here. A year ago every lab raced to ship their most capable model the moment it was ready. Now Anthropic is shipping the second-best one on purpose and openly saying the top one found thousands of zero-day vulnerabilities in testing, so only 50 banks and partners get to touch it. With California's SB 53 live and the EU's Code of Practice hitting enforcement in August, expect more "we built it but you can't have it" launches. The frontier is splitting into a public tier and a vault tier, and where your model sits is becoming a policy decision, not a product one.
Here's the 5-step setup for non-technical users:
Go to claude.ai/code/routines and click "New routine."
Name it (e.g., "Morning email triage").
Write the prompt like an SOP. Routines run hands-off, so be more precise than a normal Claude chat. Example: "Pull my Gmail unreads. For each one, check for prior conversations with that contact. Draft a reply. Send me the drafts in Slack."
Pick a trigger: schedule (daily at 5 AM), webhook (a URL that other apps can "ping" to fire the routine automatically; e.g., a new Gmail arrives → routine runs), or API call (an API is how programs talk to each other; "call" means any script or tool you run can kick off the routine on demand from anywhere).
- Add connectors under Settings → Connectors. Sign in to Gmail, Slack, Notion, etc. (uses OAuth, the same "Log in with Google"-style flow you've clicked through a hundred times) so the routine can read and write in those tools.
Hit "Run now" to test.