How much does Claude Opus 5.5 cost?

Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, 20% below Opus 5's $5/$25. Cache reads are $0.20 per million (down 60% from $0.50), 5-minute cache writes $5, 1-hour cache writes $8, and the Batch API is 50% off. Fast mode, a research preview on the Claude API only, runs up to 2.5x faster at $8/$40. Anthropic says the combination of lower prices and fewer tokens per task cuts typical workload costs about 40% versus Opus 5.

Is Claude Opus 5.5 better than Claude Fable 5.1?

On Anthropic's own table Opus 5.5 beats Fable 5.1 on every listed benchmark: Terminal-Bench 4.0 66.4% vs 55.8%, FrontierCode v1.1 54.4% vs 50.3%, CursorBench 4.0 57.8% vs 51.8%, GDPval-AA v2.1 1846 vs 1735 Elo, OSWorld 2.0 81.8% vs 80.7%, and Humanity's Last Exam 67.7% vs 65.6%. Artificial Analysis independently scored it 58 on its Intelligence Index, five points above Fable 5.1 and GPT-6 Astra at 53. Anthropic cautions that in daily use the gap is narrower than the scores suggest, and Fable 5.1 remains the $10/$50 flagship.

What are the breaking changes in Claude Opus 5.5?

Four changes break code that ran on Opus 5: thinking cannot be disabled (thinking type 'disabled' or a manual budget_tokens returns a 400 error; use effort instead), forced tool use via tool_choice 'any' or 'tool' is rejected, thinking blocks are bound to the model and conversation that produced them, and on the Claude API and Google Cloud the older computer_20251124 tool is replaced by the computer_toolset_20260801 toolset. Text between tool calls now also arrives in thinking blocks that are empty at the default display setting.

Quick Answer

What Is Claude Opus 5.5? Anthropic's Sep 2026 Model

Q: What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic's frontier model released September 22, 2026, the first member of the Claude 5.5 family. Anthropic says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Claude Opus 5 at default settings. It has a 1 million token context window, 128,000 max output tokens, a June 2026 knowledge cutoff, always-on adaptive thinking with a default effort of medium, and is available on the Claude API (claude-opus-5-5), Amazon Bedrock, Google Cloud, and Microsoft Foundry.

Published: September 23, 2026

The short answer

Claude Opus 5.5 is Anthropic’s new default frontier model, released September 22, 2026: Fable 5.1-class intelligence at $4/$20 per million tokens, 40% cheaper to run than Opus 5, and the first model to score 58 on the Artificial Analysis Intelligence Index. It is the first release since Dario Amodei’s “We Must Pace the Frontier” essay, ships with Fable-tier safeguards on cyber and biology, and replaces Opus 5 as the recommended Opus for long-running agentic coding and knowledge work.

At a glance

	Claude Opus 5.5
Released	September 22, 2026
Model ID	`claude-opus-5-5` (same on Google Cloud and Microsoft Foundry; `anthropic.claude-opus-5-5` on Bedrock)
Price	$4 in / $20 out per MTok; cache read $0.20; 5m cache write $5; 1h cache write $8; Batch 50% off
Fast mode	$8 / $40, up to 2.5x speed, Claude API only (research preview)
Context / max output	1M tokens / 128K (300K on Batch with the `output-300k-2026-03-24` beta)
Knowledge cutoff	June 2026
Thinking	Adaptive, always on; effort low → medium (default) → high → xhigh → max
Cache minimum	512 tokens
Speed	Output generation >30% faster than Opus 5
Platforms	Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS
Safeguards	Fable 5.1-class on cyber, biology, distillation; preserved thinking for accounts created on/after Aug 31, 2026
Next in family	Claude Sonnet 5.5 and Haiku 5.5 “in the coming weeks”

Why it matters: the price-performance step

Opus 4.5, 4.6, 4.7, 4.8 and 5 all cost $5/$25. Opus 5.5 is the first Opus price cut since November 2025, and the cache-read cut is the bigger story: agentic sessions reprocess the same context turn after turn, so 90%+ of input tokens on a long coding job hit the cache. At $0.20 per million cached tokens (5% of the input price, versus 10% on earlier Opus), that portion of the bill drops 60%.

Anthropic’s own worked examples:

HAProxy C-to-Rust rewrite: Opus 5.5 finished in 9.5 hours versus 12 for Fable 5.1, at 51% lower cost, with both passing nearly all regression tests.
200,000-line codebase audit: under 3 hours versus 20+ hours for Opus 5, using 2.5x fewer tokens.
Merger analysis with an Excel model and deck: 63 minutes versus 93 for Opus 5, at 50% lower cost.

Artificial Analysis notes one caveat: at max effort Opus 5.5 emits ~119,000 output tokens per Intelligence Index task versus ~73,000 for Opus 5 (max), so cost per task at max is roughly level with Opus 5 despite the price cut. The savings show up at the default medium effort, which sits on AA’s intelligence-versus-cost Pareto frontier along with high and xhigh.

Benchmarks (Anthropic’s launch table, September 22, 2026)

Benchmark	Opus 5.5	Fable 5.1	Opus 5	GPT-6 Astra
Terminal-Bench 4.0	66.4%	55.8%	52.3%	57.9%
FrontierCode v1.1 (Main)	54.4%	50.3%	48.0%	53.3%
CursorBench 4.0	57.8%	51.8%	46.6%	—
GDPval-AA v2.1 (Elo)	1846	1735	1708	1542
AutomationBench (Zapier)	40.0%	31.4%	26.9%	41.4%
Humanity’s Last Exam (tools)	67.7%	65.6%	63.6%	57.2%
Terminal-Bench-Science 0.1	58.7%	52.6%	29.0%	64.6%
OSWorld 2.0 (partial)	81.8%	80.7%	74.0%	—

Independent read: Artificial Analysis measured Terminal-Bench 4.0 at 59.6% (level with GPT-6 Astra xhigh, +11 over Opus 5) and gave Opus 5.5 the lead on six of its ten Intelligence Index evaluations, including Humanity’s Last Exam (61.4%), SciCode (66.9%), GDPval-AA, AA-Briefcase (1,822 Elo, +143 over Fable 5.1) and AA-Omniscience. It trails on CritPt, AA-LCR and GDP.pdf. Anthropic’s higher HLE number uses its own harness; treat the AA figures as the cross-vendor baseline.

What changed beyond the scores

Communication. The most common Opus 5 complaint was verbose, hard-to-follow output. Opus 5.5 leads with the conclusion, uses less jargon and follows writing rules you give it; Ramp’s early tester said it “writes like a good colleague.”

Safety. It posts the best scores yet on Anthropic’s ~2,000-scenario automated behavioral audit, attempted to cross containment boundaries about 85% less often than Opus 5 or Mythos 5.1 in a new evaluation, and ties Fable 5.1 for the lowest prompt-injection success rate in Gray Swan’s benchmark. Anthropic also flags that the model “often suspects it is being evaluated,” a limitation it is open about.

Safeguards. Because Opus 5.5 matches Mythos 5.1 in biology and cybersecurity, it inherits Fable 5.1’s safeguards: most offensive-security tasks silently fall back to Opus 4.8, and biology and frontier-LLM-development tasks fall back to Opus 5. Vetted labs can apply to the Life Sciences Verification Program; the Cyber Verification Program expands to Opus 5.5 in the coming weeks.

Subscriptions. Five-hour usage limits rise on Pro, Max, Team and seat-based Enterprise plans, and subscribers get a saveable rate-limit reset. This partly reverses the September 14 weekly-limit cut.

Breaking changes for developers

Four things that worked on Opus 5 return 400 errors on Opus 5.5:

thinking: {"type": "disabled"} or a manual budget_tokens — adaptive thinking is always on; steer cost with output_config.effort.
tool_choice of any or a named tool — use auto with strict: true, or structured outputs.
Replaying a thinking block after the system prompt or tools changed (enforced by default for accounts created on or after August 31, 2026).
The legacy computer_20251124 tool on the Claude API and Google Cloud — move to computer_toolset_20260801 (Bedrock still accepts the old tool).

A fifth change is silent: text between tool calls now arrives inside thinking blocks whose text is empty at the default display setting, so progress streams go quiet unless you set display. See How to migrate to Claude Opus 5.5.

How it fits the September 2026 lineup

Opus 5.5 lands the same day OpenAI halved GPT-6 Sol and Luna prices, one day after xAI’s Grok 4.7 and Xiaomi’s MiMo-V2.6. The frontier tier ($10/$50: Fable 5.1, GPT-6 Astra) is untouched; the price war is in the tier below, where Opus 5.5 at $4/$20 now sits directly against GPT-6 Sol at $2/$10. For that matchup see Claude Opus 5.5 vs GPT-6 Sol; for the upgrade decision from Opus 5, see Opus 5.5 vs Opus 5.

Last verified: September 23, 2026.