Claude Opus 5.5 (Sep 22, 2026): pricing $4/$20, always-on thinking, API breaks, Claude Code 2.1.280
Sep 22 launch: Opus 5.5 at $4/$20 with $0.20 cache reads; thinking always-on and tool_choice any/tool → 400; Claude Code 2.1.280 defaults Pro/Team Standard to Opus.
On Sep 22, 2026, Anthropic launched Claude Opus 5.5 (claude-opus-5-5): $4 / $20 per MTok and cache reads $0.20. Adaptive thinking is always on (disable → 400); tool_choice any/tool → 400; computer use on API/GCP needs computer_toolset_20260801. Claude Code 2.1.280 makes it the default Opus, moves Pro/Team Standard default to Opus, and drops pre-per-model saved effort on newly released models. Five-hour limits ↑ plus a savable usage-limit reset ([GAP] exact caps). Thin same-day N : see rate limits, max effort, prior 2.1.273-278.
Pricing and limits
Per Newsroom and Platform RN (Sep 22):
| Per 1M tokens | Opus 5.5 | Opus 5 |
|---|---|---|
| Input | $4 | $5 |
| Output | $20 | $25 |
| Cache reads | $0.20 | $0.50 |
| Cache writes | $5 | $6.25 |
Newsroom attributes “~40% less on typical workloads” to Anthropic’s own tests : not a CF measurement. Fast mode research preview is listed at $8 / $40 (Newsroom); not expanded here.
Newsroom also says five-hour usage limits increased on Pro, Max, Team, and seat-based Enterprise, and subscription users get a rate-limit reset they can save and use when chosen. Exact new caps and reset UI path are [GAP] in primary pages checked 2026-09-23 : do not invent numbers; operator playbook stays on How do I manage Claude Code rate limits?.
API breaks operators hit first
Platform RN Sep 22 + What’s new in Opus 5.5:
- Thinking always-on.
thinking: {"type": "disabled"}orenabled+budget_tokens→ 400. Omitthinkingor sendadaptive. Steer depth with effort (defaultmediumon Opus 5.5). - Forced tools.
tool_choicetypesany/tool→ 400. Useauto(with strict tool use for schema-valid inputs) ornone. - Computer use. On Claude API and Google Cloud, require
{"type":"computer_toolset_20260801"};computer_20251124→ 400. On Amazon Bedrock,computer_20251124still works (Platform RN platform split).
Keep migration thin : follow Anthropic’s migration guide; no full agent-loop rewrite here.
Claude Code 2.1.280
Raw CHANGELOG head is ## 2.1.280. There is no ## 2.1.279 between 2.1.278 and 2.1.280.
Must lines (CHANGELOG 2.1.280 only):
- Added Claude Opus 5.5 (
claude-opus-5-5), now the default Opus : 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads. - Changed default model on Pro and Team Standard from Sonnet to Opus (matches Max / Team Premium / Enterprise).
- An effort level saved before
/effortbecame per-model no longer applies to newly released models such as Opus 5.5; they start at default until you pick a level : see How do I set Claude Code max effort level?.
Dense UI/fix/VS Code lists in 2.1.280 stay in the CHANGELOG : not expanded here. Prior catch-up ending at 2.1.278: Claude Code 2.1.273-278 (superseded HEAD).
Upgrade readout
| Operator | Action |
|---|---|
| API / GCP integrators | Drop thinking-off and forced tool_choice; migrate computer use to computer_toolset_20260801 |
| Bedrock computer-use | computer_20251124 still OK on Opus 5.5 (RN) |
| Claude Code | Upgrade to ≥ 2.1.280; check /model and /effort after Opus default |
| Pro / Team Standard | Expect default model Opus (claude-opus-5-5) |
Sources (checked 2026-09-23)
- Anthropic Newsroom : Claude Opus 5.5 : Sep 22 launch, $4/$20, cache reads $0.20, five-hour limits ↑, savable rate-limit reset
- Platform release notes : September 22, 2026: model id, thinking / tool_choice / computer_toolset 400s, Bedrock keep
- What’s new in Claude Opus 5.5 : exact 400 messages, effort default
medium - Claude Code CHANGELOG.md :
## 2.1.280only (no 2.1.279 between 278 and 280) - Live: /guides/how-do-i-manage-claude-code-rate-limits · /guides/how-do-i-set-claude-code-max-effort-level · /guides/claude-code-2-1-273-278-2026-09-19