Model
Model explorer

Claude Opus 5.5

CLOSED
Anthropic · Claude 5.5 family · released Sep 22, 2026

Model id claude-opus-5-5, released 2026-09-22 as 'the first model in our new Claude 5.5 family'. Anthropic says it 'performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5'. Priced at $4/$20 per MTok (20% below Opus 5), with cache reads at $0.20 (60% less than Opus 5). It has a 1M context, 128K max output, adaptive thinking that is always on (it can't be disabled), and medium default effort. It ships with Fable-5.1-class cyber/bio safeguards, and flagged cyber tasks fall back to Opus 4.8. Launch scores were run with safeguards on, which Anthropic says likely depresses some rows.

ReasoningCodingVisionFunction callingTool useAgentic
3266.5
Elo · rank #1
Parameters
Undisclosed
Active params
—
Context
1M tokens
Architecture
Proprietary model; architecture and parameter count undisclosed; adaptive thinking with a runtime effort parameter (low to max), 1M-token context
License
Proprietary
Languages
—
API price (in/out)
$4 / $20
Modalities
text · vision
Benchmark results
Bar shows position within the tracked field; marker = field best
AutomationBenchAgents40.0%#11
best: DeepSeek-V4.1-Flash · 54.8%
ChartographyVision64.4%#2
best: Gemini 4 Argon · 71.6%
CursorBench 4.0Coding57.8%#1
best: this model · 57.8%
DeepSWECoding74.2%#3
best: Gemini 4 Argon · 77.9%
best: this model · 65.3%
FrontierCodeCoding54.6%#1
best: this model · 54.6%
FrontierSWE v2Coding62.3%#1
best: this model · 62.3%
HealthBench ProfessionalKnowledge65.6%#1
best: this model · 65.6%
Humanity's Last ExamReasoning67.7%#1
best: this model · 67.7%
OfficeQA ProAgents67.7%#2
best: Claude Fable 5.1 · 69.0%
OSWorld 2.0Agents81.8%#1
best: this model · 81.8%
best: this model · 93.9%
best: this model · 61.4%
SWE-bench ProCoding89.9%#1
best: this model · 89.9%
Terminal-Bench 4.0Agents66.4%#2
best: Claude Sonnet 5.5 · 70.6%
best: GPT-6 Astra · 64.6%
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
$4
Output / M tok
$20
API price $4/$20 · each benchmark row carries its own source badge (see methodology)