On November 24, 2025, Anthropic officially announced Claude Opus 4.5. Optimized for coding, agents, and computer use, this model has become one of the most notable models in the AI industry as of January 2026.
This analysis is based on authoritative sources including Anthropic's official announcement, The New Stack, and AI Business.
Key Achievement: SWE-bench 80.9%
Claude Opus 4.5 achieved 80.9% on SWE-bench Verified, becoming the first model to break the 80% barrier. SWE-bench measures the ability to solve real GitHub issues, evaluating practical software engineering capabilities.
"Anthropic administered to Opus 4.5 a performance engineering test used in its actual hiring process. In this 2-hour test, Opus 4.5 scored higher than any human candidate who has ever applied."
- Anthropic Official Page
Key Benchmark Performance Comparison
Key benchmark results as reported by AI Business and Business Analytics:
- SWE-bench Verified: 80.9% (Industry first to break 80%)
- OSWorld: 66.3% (#1 on computer use benchmark)
- Pricing: 66% reduction compared to previous Opus
Revolutionary "Effort" Parameter
According to Anthropic's announcement, Opus 4.5's most notable feature is the "effort" parameter. This API feature (currently in public beta) allows developers to control the model's reasoning depth.
- Low effort: Fast and cost-efficient responses
- Medium effort: Sonnet 4.5-level performance, 76% fewer output tokens
- High effort: Optimized for complex multi-step reasoning
The New Stack describes this as a "reasoning knob" that allows developers to balance cost and performance based on the situation.
Pricing and Availability
Pricing according to Microsoft Azure Blog and Anthropic's official announcement:
- Input tokens: $5 / million tokens
- Output tokens: $25 / million tokens
- Prompt caching: Up to 90% cost savings
- Batch processing: 50% cost savings
Platform Availability
- Claude Developer Platform (default)
- Amazon Bedrock
- Google Cloud Vertex AI
- Microsoft Foundry (public preview)
- GitHub Copilot paid plans
- Microsoft Copilot Studio
Shifting Industry Landscape
According to The New Stack, the emergence of Opus 4.5 has once again shaken up the competitive dynamics among the AI Big 3.
"A week before the Opus 4.5 announcement, Google briefly claimed the performance throne with Gemini 3 Pro. However, Opus 4.5's specialized reasoning and coding capabilities have put Anthropic back on top."
- The New Stack
Current AI Big 3 Comparison (January 2026)
- OpenAI GPT-5.2: First to break ARC-AGI 90%+, AIME 2025 100% (OpenAI Official)
- Google Gemini 3 Pro: Maintains #1 on LMArena leaderboard
- Anthropic Claude Opus 4.5: SWE-bench 80.9%, coding/agent specialized
Cowork: New Computer Use Feature
According to TechCrunch, Anthropic also unveiled a research preview called Cowork. This feature allows Claude to directly access local folders on a user's computer to complete multi-step tasks.
- Available for Max and Pro subscribers
- Local file system access
- Multi-step autonomous task execution
Usage Limit Issues
The Register reports that Claude Code users have raised concerns about usage limit changes. Some users noted that token allocations are being consumed quickly.
Anthropic doubled usage limits during the holiday period from December 25-31, 2025, to utilize idle computing capacity.
Anthropic's Strategy: "Do More With Less"
In a CNBC interview, Anthropic co-founder Daniela Amodei explained the company's strategy.
"The next step isn't about winning with the biggest pre-training. The one who delivers the most capability per compute dollar wins."
- Daniela Amodei, Anthropic Co-founder
While OpenAI announces a $1.4 trillion compute investment, Anthropic is taking a measured approach to spending and algorithmic efficiency.
2026 Revenue Outlook
- Anthropic: 2025 $4.7B → 2026 target $15B
- OpenAI: 2025 $13B+ → 2026 target $30B
Implications for Developers
Here's what Claude Opus 4.5's arrival means for developers:
- Coding tasks: Highest success rate in real development work with SWE-bench 80.9%
- Cost optimization: Balance cost-performance for each situation with the effort parameter
- Agent workflows: Top-tier autonomous task execution with OSWorld 66.3%
- Multi-platform: Available on AWS, GCP, and Azure
Future Outlook: Awaiting Claude 5
According to The New Stack, the industry expects first news about Claude 5 in spring 2026. Until then, the main focus is how Opus 4.5's new reasoning capabilities will be integrated into business.