Claude Opus 4.8 Landed Today — Here’s What Changes on Your Mac

Anthropic shipped Claude Opus 4.8 this morning. If you only read the headline you’d file it under “incremental point release and move on,” and the company itself frames it that way — a modest but tangible step up from Opus 4.7, available at the same price. But a few of the things riding alongside it actually matter for how I work day to day on a Mac, and they’re worth ten minutes of your attention before you go back to whatever you were building.

Here’s the short version of what dropped, and then where it touches the tools most of us in the Apple ecosystem already have open.

The model itself

Opus 4.8 improves on 4.7 across coding, agentic, and reasoning benchmarks, and Anthropic is positioning it as a sharper collaborator rather than a raw-horsepower jump. The line that caught my eye wasn’t a benchmark, though — it was the honesty work. Anthropic says 4.8 is roughly four times less likely than its predecessor to let flaws in code it wrote slip by without comment, and early testers report it flags its own uncertainty more readily instead of confidently declaring victory on a half-finished task.

If you’ve ever had a coding agent tell you everything’s working and then watched the build fail, you understand why that’s the upgrade I care about most. An agent that says “I’m not sure this handles the edge case” is worth more than one that’s cheerfully wrong.

Pricing is unchanged: $5 per million input tokens and $25 per million output on regular usage. The model string is claude-opus-4-8 if you’re calling it through the API.

Effort control — the one most people will actually use

This is the change that lands directly in the Claude app on your Mac and iPhone. There’s now a control next to the model selector that lets you pick how much effort Claude puts into a response, and it’s available on every plan including the consumer tiers.

Higher effort means Claude thinks more often and more deeply before answering. Lower effort means faster responses that burn through your rate limits more slowly. Opus 4.8 defaults to high, which Anthropic considers the best balance — but the practical takeaway for anyone working off a laptop on the move is that you now have a dial. Quick reformatting job or a throwaway question? Drop the effort, save your limits. Architecture decision you’re going to live with for six months? Crank it.

Claude Code on macOS: dynamic workflows

For those of us who live in Terminal, the bigger news is a research-preview feature called dynamic workflows in Claude Code. It lets Claude plan a large task, spin up hundreds of parallel subagents in a single session, and verify its own output before reporting back. The example Anthropic gives is a codebase-scale migration across hundreds of thousands of lines — kickoff to merge, with your existing test suite as the bar it has to clear.

The catch: dynamic workflows is limited to Claude Code for Enterprise, Team, and Max plans, so it’s not on the Pro tier yet. Fast mode is the other Claude Code note — Opus 4.8 can run at 2.5× speed, and fast mode is now three times cheaper than it was on previous models, at $10 / $50 per million tokens.

VSCode, Cursor, Codex, and Raycast users

A few signals from the launch quotes are relevant if Claude isn’t your only tool:

• Cursor reported that Opus 4.8 beats prior Opus models at every effort level on their CursorBench, with more efficient tool calling — fewer steps for the same result. If you run Claude as the model behind Cursor on your Mac, that efficiency shows up as lower token spend on the same work.

• On Terminal-Bench 2.1, Anthropic notes GPT-5.5 scores 83.4% using the Codex CLI harness, so if you’re bouncing between Codex and Claude in the terminal, the gap is closer than the marketing on either side suggests — test both on your own repo before committing.

• Raycast doesn’t get a direct mention, but the effort control is the thing to watch. Raycast AI Commands that route to Opus will inherit the same speed-versus-depth tradeoff, so a command you fire fifty times a day is a good candidate for a lower effort setting once that propagates.

For developers building on the API

One quiet but useful change: the Messages API now accepts system entries inside the messages array. You can update Claude’s instructions mid-task — permissions, token budgets, environment context — without breaking the prompt cache or faking a user turn to do it. If you’re building an agent harness on a Mac and you’ve been hacking around this, it’s a clean fix.

Worth doing this week

Open the Claude app, find the new effort control, and spend a day noticing when you actually want it high versus low — it changes how fast you burn limits more than any setting they’ve shipped so far. If you’re on Max or Team and you run Claude Code, dynamic workflows is worth a real test on something you’d normally break into a dozen manual steps.