Claude Opus 4.5, released November 23, 2025, is Anthropic’s most intelligent AI model, achieving a record 80.9% on SWE-bench Verified while costing 66% less than its predecessor at $5/$25 per million tokens. The model handles autonomous coding sessions for 20-30 minutes, introduces an effort parameter for token efficiency, and delivers step-change improvements in reasoning, coding, and complex problem-solving. It outperforms GPT-5.1 and Gemini 3 Pro on coding benchmarks and includes new integrations with Chrome and Excel.
Claude Opus 4.5 is Anthropic’s newest flagship AI model, setting a new standard for coding, autonomous agents, and computer use while being significantly more affordable than previous Opus models. Released on November 23, 2025, it achieves state-of-the-art performance on software engineering benchmarks and introduces practical features that make advanced AI capabilities accessible to more developers and enterprises.
What Is Claude Opus 4.5?
Claude Opus 4.5 represents Anthropic’s most intelligent model to date, combining maximum capability with practical performance. The model delivers flagship intelligence at $5 per million input tokens and $25 per million output tokens, a 66% price reduction from Claude Opus 4.1’s $15/$75 pricing. This dramatic cost decrease makes enterprise-grade AI accessible for everything from startup projects to large-scale deployments.
The model uses the identifier claude-opus-4-5-20251101 and is available through Claude’s API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. It features a 200,000-token context window and supports all major programming languages with exceptional performance across seven out of eight tested languages on SWE-bench Multilingual.
Key Features and Capabilities
Autonomous Coding for 20-30 Minutes
One of Claude Opus 4.5’s breakthrough capabilities is its ability to code autonomously for extended periods without human intervention. Adam Wolff from Anthropic shared that the model “could routinely code autonomously for 20 to 30 minutes,” often completing complex tasks idiomatically and correctly. This represents a fundamental shift toward AI handling long-horizon programming tasks that previously required constant developer supervision.
Early testers consistently reported that Claude Opus 4.5 “just gets it,” handling ambiguity and reasoning about tradeoffs without hand-holding. When pointed at complex, multi-system bugs, the model figures out fixes that were “near-impossible for Sonnet 4.5 just a few weeks ago”.
Effort Parameter for Token Control
Claude Opus 4.5 is the only model supporting the new effort parameter, which lets developers control response thoroughness and token efficiency with a single setting. The parameter offers three levels:
- High effort: Maximum thoroughness for complex analysis
- Medium effort: Balanced approach for production use cases
- Low effort: Most token-efficient responses for high-volume automation
At medium effort, Opus 4.5 matches Sonnet 4.5’s best SWE-bench score while using 76% fewer output tokens. At maximum effort, it exceeds Sonnet 4.5 by 4.3 percentage points while still using 48% fewer tokens.
Enhanced Computer Use with Zoom Action
The model introduces a new zoom action that enables detailed inspection of specific screen regions at full resolution. This allows Claude to examine fine-grained UI elements, small text, and detailed visual information that might be unclear in standard screenshots. The zoom capability proves particularly valuable for inspecting small UI controls, reading fine print, analyzing dense interfaces, and verifying precise visual details before taking actions.
Thinking Block Preservation
Claude Opus 4.5 automatically preserves all previous thinking blocks throughout conversations, maintaining reasoning continuity across extended multi-turn interactions and tool use sessions. This ensures the model effectively leverages its full reasoning history when working on complex, long-running tasks, a critical feature for maintaining context in hours-long coding sessions.
Claude Opus 4.5 Benchmarks and Performance
SWE-Bench Verified Results
Claude Opus 4.5 achieved a record-breaking 80.9% on SWE-bench Verified, the industry standard for real-world software engineering tasks. This score surpasses all competing models including GPT-5.1 (77-78%) and represents the highest performance ever recorded on this benchmark. The model also scored 89.4% on Aider Polyglot coding problems, significantly beating Sonnet 4.5’s 78.8%.
In Anthropic’s internal testing, Claude Opus 4.5 scored higher on a notoriously difficult performance engineering take-home exam than any human candidate within the prescribed 2-hour time limit. This result demonstrates the model’s ability to assess technical ability and judgment under time pressure at a level that matches or exceeds strong human candidates.
Performance vs GPT-5.1 and Gemini 3 Pro
Comparative testing shows Claude Opus 4.5 excels in agentic coding and computer use while maintaining competitive performance across reasoning and general intelligence tasks. On SWE-bench Verified, Claude leads with 80.9% compared to GPT-5.1’s 77-78%. For algorithmic tasks, GPT-5 uses approximately 90% fewer tokens, but Claude delivers more thorough reasoning with comprehensive comments and test cases.
Claude Opus 4.5 demonstrates superior performance in debugging and long-horizon coding tasks that require “integrating understanding across a project”. Reddit users testing the model report it comprehends instructions on the first attempt and produces “organized, coherent, and remarkably sharp” work without needing repetitive explanations.
Pricing and Availability
Claude Opus 4.5 is available at $5 per million input tokens and $25 per million output tokens across Anthropic’s API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. AWS Bedrock and Google Vertex offer global endpoints at standard pricing and regional endpoints with a 10% premium for guaranteed geographic data routing. The Claude API remains global-only with no regional pricing variations.
This pricing represents exceptional value compared to competitors: Gemini 3 Pro costs $2/$12 (or $4/$18 for contexts exceeding 200K tokens), while Claude Opus 4.1 previously cost $15/$75. For users requiring complete implementations on the first try, the higher cost per token often proves worthwhile due to reduced iteration cycles.
Claude Opus 4.5 vs Competitors
| Feature | Claude Opus 4.5 | GPT-5.1 | Gemini 3 Pro |
|---|---|---|---|
| SWE-Bench Verified | 80.9% | 77-78% | ~75% |
| Input Pricing | $5/MTok | ~$10/MTok | $2-4/MTok |
| Output Pricing | $25/MTok | ~$30/MTok | $12-18/MTok |
| Context Window | 200K tokens | 400K tokens | 200K+ tokens |
| Autonomous Coding | 20-30 min sessions | 10-15 min sessions | 10-15 min sessions |
| Effort Parameter | Yes | No | No |
| Zoom Action | Yes | No | No |
Real-World Use Cases
Software Development and Code Migration
Claude Opus 4.5 excels at production-quality code generation, refactoring, and modernization across eight programming languages. Developers report 50-75% reductions in both tool calling errors and build/lint errors compared to other models. The model handles code migration by analyzing legacy systems, planning migration strategies, and executing changes while maintaining functionality.
AI Agents and Multi-Tool Orchestration
The model demonstrates breakthrough capabilities in multi-tool orchestration, seamlessly coordinating hundreds of tools in complex workflows. In testing by Rakuten, agents achieved peak performance in just 4 iterations, while other models couldn’t match that quality after 10 attempts. This self-improving capability makes Claude Opus 4.5 ideal for cybersecurity operations, financial analysis, and end-to-end software engineering workflows requiring 10+ tools.
Excel and Chrome Automation
Claude Opus 4.5 sets new standards for Excel automation and financial modeling, with 20% improved accuracy and 15% better efficiency on Anthropic’s internal evaluations. Through Claude for Chrome, the model accesses the browser’s Document Object Model to read documentation, navigate multi-step flows, fill forms, and connect disparate web interfaces.
Technical Specifications
Model Identifier: claude-opus-4-5-20251101
Context Window: 200,000 tokens
Pricing: $5 input / $25 output per million tokens
Release Date: November 23, 2025
Programming Languages: Python, JavaScript, TypeScript, Java, C++, Go, Rust, PHP
Key APIs: Effort parameter, zoom action, thinking block preservation, memory tool
Platforms: Claude API, AWS Bedrock, Google Vertex AI, Microsoft Foundry
Safety: Most robustly aligned model with industry-leading prompt injection resistance
Pros and Cons
Pros:
- Record-breaking 80.9% on SWE-bench Verified—highest among all AI models
- 66% price reduction from Opus 4.1 while delivering superior performance
- Autonomous coding for 20-30 minutes without human intervention
- Effort parameter enables precise control over token usage and thoroughness
- Enhanced computer use with zoom action for detailed UI inspection
- Excels at long-horizon tasks with consistent performance across extended sessions
- Industry-leading safety and alignment with robust prompt injection resistance
- Comprehensive platform availability across major cloud providers
Cons:
- Higher per-token cost than Gemini 3 Pro ($5 vs $2 input, $25 vs $12 output)
- Smaller context window than GPT-5.1 (200K vs 400K tokens)
- Uses more tokens than GPT-5 for algorithmic tasks (though with more thorough reasoning)
- Some benchmarks lag behind competitors in visual reasoning and multilingual Q&A
- Effort parameter and advanced features exclusive to Opus 4.5, not available in Sonnet/Haiku tiers
Source: Anthropic
