The default is Opus

Pro and Team Standard just started on Opus. Nobody asked them in the prompt.

Claude Code 2.1.280, dated September 22, 2026 in AI/TLDR’s release note and confirmed in Anthropic’s Claude Code changelog, does two load-bearing things in one ship. First, it adds Claude Opus 5.5 (claude-opus-5-5) and makes it the default Opus model: 1M context, list prices of $4 input and $20 output per million tokens, $0.20 per million for cache reads. Second, it changes the default model on Pro and Team Standard plans from Sonnet to Opus, matching Max, Team Premium, and Enterprise. Anthropic’s September 22 Opus 5.5 launch post is the model context: lower token prices than Opus 5, higher five-hour usage limits on Pro, Max, Team, and seat-based Enterprise, and thinking that you cannot turn off on Opus 5.5.

This is a primary-source review. No hands-on theater. Distinct from Vault’s classifier-billing piece on 2.1.278, the projects-as-runtime note, the Codex voice review, and the one-Claude workspace work. The question: who got a heavier default without touching /model, what that does to quotas and subagents, and which permission fix matters.

What it is

Claude Code is Anthropic’s terminal coding agent. 2.1.280 is a model-default release with a permission fix.

Opus 5.5 is the new model ID. Anthropic’s launch post prices it at $4 / $20 per million tokens with cache reads at $0.20, down from Opus 5’s $5 / $25 and $0.50 cache reads. Fast mode is listed at $8 / $40. The model-config docs say Opus 5.5 requires Claude Code v2.1.280 or later, that the opus alias on the Anthropic API now resolves to Opus 5.5, and that before 2.1.280 the default option resolved to Sonnet 5 on Pro and Team Standard and to Opus 5 on Max, Team Premium, Enterprise, and API paths.

The plan flip matters more than the model marketing. Mixed-News (September 23) restates the changelog cleanly: Pro at $20 per month in the US ($17 annual) and Team Standard at $25 monthly ($20 annual) now open on an Opus model unless the developer picks another one. Those are the two cheapest paid tiers. Max, Team Premium, and Enterprise already defaulted to Opus. 2.1.280 flattens that split.

What changed

Move one: Opus 5.5 becomes the Opus family tip. Changelog: “Added Claude Opus 5.5 (claude-opus-5-5), now the default Opus model.” Model-config: opus resolves to Opus 5.5 on Anthropic API, Claude Platform on AWS, Bedrock, and Google Cloud’s Agent Platform. Anyone on opus, default (post-flip), or an unset session default on the newly aligned plans gets 5.5.

Move two: Pro and Team Standard default to Opus. Changelog: “Changed the default model on Pro and Team Standard plans from Sonnet to Opus, matching Max, Team Premium, and Enterprise.” Model-config’s default table now lists Pro, Max, Team, Enterprise, and Anthropic API as defaulting to Opus 5.5, with a before-v2.1.280 note that Pro and Team Standard used to land on Sonnet 5. AI/TLDR and Mixed-News both treat this as the product story.

Move three: symlink writes stop getting a free pass. Changelog: writes through a symlinked path used to be judged by their in-tree spelling. The prompt now names where the write lands, and acceptEdits, allow rules, and auto mode no longer approve a write that resolves outside the project. AI/TLDR: the path as typed could pass while the file landed elsewhere.

Move four: effort no longer blindly carries onto new models. An effort level saved before /effort became per-model no longer applies to newly released models such as Opus 5.5; they start at their own default (medium for 5.5 per model-config) until you pick a level. Opus 4.7, Opus 4.8, and Fable 5 also stop holding launch-default effort over levels set in -p, the Agent SDK, or settings effortLevel.

Move five: auto mode stops retry-spinning on declined safety review. A declined safety review is denied once with a note that retrying will not help; unanswered checks back off and stop after ten. Adjacent to last week’s classifier-billing story, not a replacement for it.

What works

For teams that already wanted Opus on every seat, the Pro / Team Standard flip removes a quiet class split. Standard and Pro no longer start thinner than Max or Premium by default.

For API and cloud accounts, Opus 5.5’s list pricing is the efficiency story: launch post cites about 40% less cost than Opus 5 on typical workloads, faster output, and cache reads at $0.20 instead of $0.50. That is Console / provider billing, not a Pro seat invoice.

For platform owners, the symlink fix is the adult change. After 2.1.280, the prompt names the real destination and auto / allow no longer bless an out-of-project resolve. Re-audit symlink-heavy trees that link build outputs or secrets into the project.

Thinking always on for Opus 5.5 is documented, not optional. Model-config: you cannot turn thinking off on Opus 5.5 or the Fable models. Depth is /effort, not a thinking off switch. Costs docs: thinking tokens bill as output tokens. On subscription that hits allowance. On API that hits the meter. Opus 5.5 starts at effort medium until you set it; old pre-per-model effort does not carry over.

What breaks (or gets expensive)

Silent Opus on cheap seats. A developer who never touched /model wakes up on Opus 5.5. Subagents that inherit the session model inherit Opus too (model-config; Mixed-News). A long agentic run that used to sit on Sonnet now draws from the Opus family quota and the shared five-hour / weekly windows with thinking always on.

Opus limit versus session limit. Costs docs: after “You’ve hit your Opus limit,” switching to a model outside that family with /model keeps you working. After a session or weekly limit, switching models does not help. Mixed-News restates the same split. Train people which ceiling fired before they “just wait.”

API spend if Opus is left as default. Costs docs name “Opus left as the default model” as one of the two usual causes of unexpectedly high spend on API or cloud-provider plans. Defaults are cost policy. Anthropic’s own guidance still reads that Sonnet handles most coding tasks well and costs less than Opus; reserve Opus for complex architectural decisions or multi-step reasoning (costs docs; Mixed-News).

Subscription users who confuse list prices with their bill. The $4 / $20 figures are API rates. Costs docs: Pro and Max subscribers have usage included, so the session cost figure is not relevant for billing. On Team and Enterprise, usage inside the seat allowance is not metered in dollars. Mixed-News: a Pro, Max, Team, or Enterprise subscriber is never invoiced those token rates for included seat usage. Cost shows up as allowance burn, then usage credits if enabled.

Symlink workflows that depended on the bug. If a team used in-tree symlink spelling to write outside the project under acceptEdits or auto mode, that path is closed. Good. Also a break ticket if someone built a workflow on the hole.

Geometric ink schematic of a routing path that forks with an oxblood upper lane.

Who it is for

Primary: Eng leads and platform owners running Claude Code on Pro or Team Standard seats who have not pinned a model, plus security owners of repos that use symlinks into or out of the working tree.

Secondary: API / Bedrock / Agent Platform / Foundry fleets adopting claude-opus-5-5 who need pricing and thinking-always-on facts without confusing them with seat economics.

Already covered: Max, Team Premium, and Enterprise users whose default was already Opus. Confirm 2.1.280+ so the Opus alias is 5.5.

Not the buyer: Anyone shopping “free Opus for $20” as unlimited heavy agent time. Limits still bind; Anthropic published no new five-hour figures in the launch post (Mixed-News).

Pricing and limits

Two meters. Do not merge them in a finance Slack.

API / Console / cloud: Opus 5.5 list is $4 input, $20 output, $0.20 cache reads per million tokens (changelog; launch post; AI/TLDR). Fast mode $8 / $40 (launch post). Billed per token.

Subscription seats: Pro, Max, Team, Enterprise draw from included allowances (rolling five-hour and weekly windows on Team / Enterprise per-seat; Pro framed as at least 5x Free per five-hour session on pricing language Mixed-News cites). Usage inside the seat allowance is not a $4/$20 invoice line (costs docs). Hit Opus family limit: switch model. Hit session or weekly limit: wait, use a saved reset if you have one, or burn usage credits if enabled.

Anthropic’s launch post promises higher five-hour limits and a savable rate-limit reset, with no published magnitudes as of Mixed-News’s September 23 write-up.

Second-order effects

Seat economics compress. Finance will ask why Opus burn rose on the cheap tier the week after 2.1.280. Answer with the changelog URL, not a vibes thread. Do not call this a free upgrade, treat $4/$20 as a Pro invoice, or merge it with 2.1.278’s classifier-billing change. Fable is still never the account-type default (model-config).

Permission culture tightens. Symlink resolution in the prompt makes allow rules honest again. Model policy becomes explicit or it becomes accident: pin Sonnet in /config, managed settings, or ANTHROPIC_DEFAULT_MODEL if daily coding should stay light, or accept that Pro and Team Standard now inherit Opus. Lead your FAQ with the default check, the limit split, and the 2.1.280+ upgrade floor.

Macro of interlocking blank metal chain links on concrete with one plate lifted.

What to do this week

  1. Upgrade Claude Code to 2.1.280 or later. Opus 5.5 and the plan-default change require it (model-config). Confirm with claude --version.
  1. On a fresh Pro or Team Standard session, run `/model` and `/status` before any real work. Record what default resolves to. If the seat should stay on Sonnet for daily coding, set Sonnet in /config or save it with /model deliberately (changelog; Mixed-News; costs guidance).
  1. Treat Opus family limits and session/weekly limits as different tickets. Opus limit means switch with /model; session or weekly limit means wait, reset, or usage credits (costs docs).
  1. Re-read allow rules and `acceptEdits` habits on symlink-heavy projects. Confirm permission prompts name the real destination. Do not approve out-of-project resolves that used to sneak through (changelog; AI/TLDR).
  1. Set effort explicitly for Opus 5.5. Do not assume an old saved level carried over. Start from the documented medium default, then raise for architecture work (model-config; changelog).
  1. Separate API cost reviews from seat allowance reviews. API accounts: watch Opus-as-default spend. Subscription seats: watch five-hour and weekly bars, not the Session dollar figure (costs docs; Mixed-News).
  1. Update the eng FAQ from primaries. Link the changelog, model-config, costs page, and the September 22 Opus 5.5 post. One paragraph: Pro/Team Standard now default to Opus 5.5; thinking always on; pin Sonnet if needed; symlink writes name the real path; $4/$20 is API list, not seat invoice.

Sharp close

The default is Opus. On Pro and Team Standard that used to mean Sonnet, the client now opens on the heavy model with thinking you cannot disable, and subagents that inherit the session follow it. Opus 5.5 is a real model drop with better list economics on the API path. The operator move this week is still boring: check /model, pin what your seats should actually run, audit symlink allows, and know which quota wall you hit. If your leadership slide only says “everyone got Opus now,” you marketed the default and skipped the controls that decide whether that default burns the week before Thursday.