Listen

Description

Anthropic quietly shipped Claude Opus 4.7 last week, and for the first time it feels like a coding model you can actually hand real work off to — not babysit. We break down what's new, what's hype, and what's worth changing in your workflow today.

In this episode

00:00 — Cold open: why 4.7 is different from a routine bump

00:30 — The three coding upgrades that matter: self-verification, longer autonomous runs, and better tool use

01:15 — SWE-bench results and the "low-effort 4.7 ≈ high-effort 4.6" claim

01:45 — Four more upgrades worth knowing

3× vision resolution

Instruction following (so good Anthropic is warning people about it)

Better cross-session memory

Higher-quality creative writing

02:45 — Pricing stays flat ($5 / $25 per million) — but mind the new tokenizer

03:05 — What else launched: Project Glass safety framework + Claude Code Max auto mode

03:25 — The bigger picture: unsupervised coding, capability shaping, and the model treadmill

03:55 — Who should actually switch today

Key takeaways

Hand-off, don't hand-hold. 4.7 is pitched as the first Claude you can leave running on multi-step coding tasks and trust the output.

Self-verification is the quiet killer feature. The model checks its own work before reporting back — fewer "looks done, actually broken" moments.

Re-tune your prompts if migrating. Instruction following got so sharp that prompts tuned for 4.6 can misfire. Loosen overly literal language.

Same price, different tokens. Dollar cost is unchanged, but the new tokenizer can use 1.0–1.35× more tokens for the same text. Watch your bills.

Vision got real. 3× resolution makes screenshot-driven agents and UI workflows meaningfully better.

Project Glass > disclaimers. Anthropic is shaping capabilities at the model level instead of bolting on warnings. Worth watching if you care about AI safety direction.

Should you switch?

Daily Claude coder: Yes. Upgrade and re-tune prompts.

Building agents: Yes — vision + memory + longer autonomy are all wins.

Casual user: No rush. 4.6 is still excellent.

Watching safety: Read the Project Glass notes.

Links

Full post: https://www.stack-snacks.com/p/opus-47-just-landed-heres-what-actually

Subscribe to the newsletter: https://www.stack-snacks.com

YouTube: Stack Snacks



This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.stack-snacks.com/subscribe