Claude Opus 5 Lands as the New Default in Claude Code - What's Different for Developers
Anthropic's new Claude Opus 5 is the default model in Claude Code, matching much of Fable 5 at ~half the price. The new /effort dial, the asks-before-guessing judgment shift, the coding benchmarks - and the honest caveats for devs.
Table of contents
Anthropic released Claude Opus 5 on July 24, 2026, and for developers the headline isn't a benchmark — it's where it shows up. Opus 5 is the new default model in Claude Code, it's included in existing plans, and Anthropic positions it as matching much of Fable 5's capability at roughly half the price. Here's what actually changes day to day, and where to keep a skeptical eye.
What Anthropic is claiming
- Long-running work. Anthropic says Opus 5 can work for hours on a task — recovering from errors and routing around blockers instead of stopping — and checks its own work as it goes. Its (vendor-reported) scores: 43.3% on Frontier-Bench v0.1 and 68.8% on DeepSWE v1.1, up from 18.7% / 59.0% on Opus 4.8.
- Judgment. It more often asks clarifying questions before guessing, pushes back on flawed instructions, and considers implications before jumping to implementation — which Anthropic frames as cleaner, higher-quality code.
- An effort dial. A new
/effortsetting trades cost for capability per task: Extra and Max push hardest (and use more of your limit), while High or below is faster and lighter for routine work. - Price. Same cost as Opus 4.8, and about half of Fable 5.
Why the "judgment" change matters more than the numbers
For working developers, the asks-before-guessing behavior is arguably a bigger deal than the score bumps. A model that stops to clarify an ambiguous ticket, or flags a flawed instruction, produces fewer confidently-wrong diffs — the failure mode that quietly eats code-review time. It pairs well with the guardrails we've argued for in AI coding governance for small teams and the visibility from agent observability: asking-before-guessing plus logged tool calls means more reviewable output.
The effort dial is the real lever
/effort is the practical knob. Use High or below for routine edits — faster and cheaper on your usage — and reserve Extra/Max for genuinely hard problems. It's the same "match the tier to the task" discipline that matters when agents run long, unattended jobs, as we covered in always-on coding agents in CI/CD. And if you keep model-specific rules, note how they interact with your setup — see AGENTS.md vs CLAUDE.md vs Cursor rules.
Honest caveats
- The benchmarks are Anthropic's own. Independent evaluations will tell the real story — we've made this point before in where GLM 5.2 actually matches Claude Opus. Treat vendor "state-of-the-art" claims with the usual skepticism until third-party tests land.
- It isn't #1 everywhere. "Matches many of Fable 5's capabilities" is not "matches all," and early coverage notes it trails other models on some specialized tasks. Right-size expectations per task.
- Cheaper ≠ free to run flat out. Half the price helps, but long-running Max-effort sessions still burn more of your limit — the effort setting is where you control that.
Bottom line
For Claude Code users this is effectively a free upgrade with a genuinely useful new dial and a behavior shift toward clarifying before coding — which should mean cleaner diffs and less review churn. Run routine work at High or below, save Max for the hard problems, and judge the leaderboard numbers once independent benchmarks confirm them.
Reporting: Anthropic, CNBC, Fortune. Benchmark figures are Anthropic's own; independent evaluations were pending at publication.
