Archive · 25 Nov 2025
Anthropic Launches Claude Opus 4.5
Anthropic, a company that makes AI, has launched a new AI model called Claude Opus 4.5.
They say it's the best in the world for tasks like coding and using computers.
It scored 80.9% on a test called SWE-bench Verified, which is higher than Google's Gemini 3 Pro and OpenAI's GPT-5.1 Codex Max.
The new model can also do better in real-world tasks, like helping a customer with a flight booking.
Anthropic also claims that this model is safer and harder to trick than previous ones.
The model is rolling out via the Claude app and website, with premium access for enterprise users starting at around $20 per month and free tiers for individual creators and hobbyists.
Anthropic launches Claude Opus 4.5, claiming it's the best for coding and computer tasks.
Opus 4.5 scores 80.9% on SWE-bench Verified, outperforming Gemini 3 Pro and GPT-5.1 Codex Max.
The model outperforms human candidates in a 2-hour technical test.
Opus 4.5 excels in real-world, multi-turn tasks, showcasing agentic AI capabilities.
Anthropic claims Opus 4.5 is its most robustly aligned model, resistant to prompt injection attacks.
- Who
- Anthropic
- What
- Launched Claude Opus 4.5 AI model
- When
- Soon after GPT-5.1 and Gemini 3 launch
- Where
- Globally
- Why
- To compete with ChatGPT and Gemini
Anthropic
Competitors (OpenAI, Google)
Model Performance
Anthropic
Claude Opus 4.5 scores 80.9% on SWE-bench Verified, outperforming Gemini 3 Pro and GPT-5.1 Codex Max.
Competitors (OpenAI, Google)
Gemini 3 Pro and GPT-5.1 Codex Max have lower scores of 76.2% and 77.9% respectively.
Agentic AI Capabilities
Anthropic
Opus 4.5 excels in real-world, multi-turn tasks, showcasing superior agentic AI capabilities.
Competitors (OpenAI, Google)
Competitors' models may struggle with complex, multi-step tasks and rigidly refuse changes in certain scenarios.
Safety and Alignment
Anthropic
Opus 4.5 is claimed to be the most robustly aligned model, resistant to prompt injection attacks.
Competitors (OpenAI, Google)
Competitors' models may be more susceptible to prompt injection attacks.
- Model Name
- Claude Opus 4.5
- SWE-bench Verified Score
- 80.9%
- Gemini 3 Pro Score
- 76.2%
- GPT-5.1 Codex Max Score
- 77.9%
- Benchmark
- τ2-bench
- Claimed Advantage
- Most robustly aligned model
Sources
Anthropic launches Claude Opus 4.5 to take on ChatGPT and Gemini: here's what you should know
Anthropic Unveils Claude Opus 4.5, Predicts AI-Led Future for Software Engineering
Anthropic Claude Opus 4.5 released: How it compares to ChatGPT 5.1 and Google Gemini 3.0
