Your Claude Code Default Just Changed to Opus 5.5. Here's What That Actually Costs You
Anthropic shipped Claude Opus 5.5 on September 22, 2026: cheaper and faster than Opus 5, matching Claude Fable 5.1 on most work. But it also quietly became the default model on the $20 Pro plan and Team Standard, replacing Sonnet. Here's what changed, what breaks, and what to check before it burns your usage window.
Two months ago I wrote about Claude Opus 5 quietly becoming the default on Max and Pro's top tier, and about the effort dial mattering more than the benchmarks. Last week Anthropic did it again, and this time the change that will actually bite you isn't in your prompts. It's in your bill and your usage window.
Claude Opus 5.5 shipped September 22, 2026. It's cheaper than Opus 5, it's faster, and Anthropic says it performs at Claude Fable 5.1 level on most work while costing a fraction of what Fable 5.1 costs. One tester's data point, straight from the announcement: a 680,000-line code migration finished in under a day, the kind of work that would have taken an engineering team weeks. That's the headline. It's not the part I'd tell a Claude Camp attendee to go check today.
What actually shipped
Opus 5.5 (claude-opus-5-5) drops input tokens from $5 to $4 per million, output from $25 to $20 per million, and cache reads from $0.50 to $0.20 per million, a 60 percent cut on the reads that make prompt caching worth doing in the first place. Anthropic's own framing puts the typical workload savings at 40 percent, with output generation running about 30 percent faster than Opus 5.
On the numbers that matter for agentic coding, Opus 5.5 jumps from 52.3 percent to 66.4 percent on Terminal-Bench 4.0, the eval closest to what Claude Code actually asks a model to do all day: run commands, read output, decide the next step. It's a real capability jump, not a repricing with a new name on it.
None of that required me to touch anything. This did.
The bigger deal: it's your default now, on the $20 plan too
Buried in the same release, Claude Code changed the default model on Pro and Team Standard plans from Sonnet to Opus, matching what Max, Team Premium, and Enterprise already had. If you're running Claude Code on the $20 Pro plan, the model answering your prompts this morning is a different one than it was last Monday, and you didn't touch a config file to get there.
The plan price didn't move. The meter did. Anthropic's own support docs put it plainly: Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku. Your 5-hour and weekly usage windows are shared across whatever model you're on, so a default swap from Sonnet to Opus means the same window drains faster per prompt, not because you're doing more work, but because each turn now costs more of the budget. If you hit a limit, /model switches you to something lighter and you keep working. That's the fix, but you have to know to reach for it.
For a solo operator running Claude Code all day, every day, this is worth five minutes: run /model right now and check what you're actually on. If it says Opus and you're doing high-volume, low-stakes work (formatting, boilerplate, straightforward refactors), that's paying Opus prices for Sonnet-shaped tasks. Drop to Sonnet for the grunt work, let Opus pick up automatically for anything that survives long enough to need xhigh effort or deep multi-file reasoning.
What breaks if you're building on the API
If you're hitting the Messages API or the Agent SDK directly rather than going through Claude Code, three things changed and none of them fail gracefully:
The thinking field can't be touched at all now, at any effort level. Opus 5 let you disable thinking at low or medium effort and only blocked it at xhigh and max. Opus 5.5 rejects both {"type": "disabled"} and {"type": "enabled", ...} with a 400, full stop. Omit the field entirely and control depth with effort instead. If your Opus 5 code explicitly set thinking, it breaks on upgrade regardless of what effort level you're running.
tool_choice types "any" and "tool" are gone. Both return a 400. Anthropic's fix is tool_choice: "auto" paired with strict tool use, which is a real behavior change if your code was forcing a specific tool call.
Computer use needs the new computer_toolset_20260801 toolset on the Claude API and Google Cloud. The old computer_20251124 tool 400s. Bedrock is the one exception, it keeps working there unchanged, which is exactly the kind of platform-specific asterisk that's easy to miss if you're testing on one provider and shipping to another.
What I'm actually checking
I grepped every place I've got a thinking field set explicitly, since Opus 5.5 doesn't care what effort level it's paired with anymore, it just rejects the field outright. I ran /model on my own daily-driver session and confirmed I'm on Opus by default now, then thought about which of my routine tasks (drip-email QA, changelog grepping, this post) actually need it versus running fine on Sonnet at a quarter of the per-turn cost. And if you're running an Agent SDK build with tool_choice forcing a specific tool, that's the one to test before you touch the model string in production.
The pricing cut is the good news. The default swap on the plan most solo builders are actually paying for is the part that changes what your month looks like if you don't go check it.
Sources: Introducing Claude Opus 5.5 · What's new in Claude Opus 5.5 / release notes · Migrating from Claude Opus 5 · Models, usage, and limits in Claude Code · Claude Code changelog
Related posts
Claude Opus 5 Shipped. The Real Change Is the Effort Dial, Not the Benchmarks
Anthropic released Claude Opus 5 on July 24, 2026: frontier-class coding at half the price of Claude Fable 5, same pricing as Opus 4.8. But the change that will actually bite your code is a default-behavior swap around thinking and effort. Here's what shipped, what breaks, and what I'm doing differently.
Claude Code's Default Model Just Changed: What Sonnet 5 Actually Gets You
Anthropic made Claude Sonnet 5 the default model in Claude Code on June 30, 2026, with a native 1M-token context window at standard pricing. Here's what actually changed, the benchmarks that matter, and what I'm doing differently this week.
Claude Code Now Coordinates Its Own Parallel Sessions. I've Been Doing That Job By Hand.
Anthropic relaunched Claude Code Projects on September 17, 2026: one conversation that starts, tracks, and reports back on parallel cloud sessions ("threads") sharing memory, instructions, and repos. Here's what it actually replaces, and what it costs you in practice.