China’s Xi speaks with Trump by phone, discusses Taiwan and bilateral ties
Investing.com -- Anthropic unveiled Claude Opus 4.5 on Monday, a major upgrade to its frontier artificial intelligence platform, delivering what the company calls “the best model in the world for coding, agents, and computer use.” Available as of today across apps, API, and major cloud platforms, the model promises heightened reasoning skills, better performance in unseen domains, and substantial improvements in real-world software engineering benchmarks.
Priced at $5/$25 per million tokens, the new offering aims to make advanced Opus-level capabilities more accessible to individuals, teams, and enterprises. Opus 4.5 leads on the SWE-bench Verified test, outperforming other frontier models in engineering problem solving, and offering new capabilities that Anthropic says will reshape how work gets done.
According to Anthropic’s internal testing, “Opus 4.5 just ‘gets it’.”
The model shows a marked improvement in agentic capabilities, solving complex tasks with human-like creativity. The model also achieved the highest score ever, surpassing any human candidate, on Anthropic’s take-home engineering test under time constraints, highlighting AI’s expanding role in skilled technical roles.
Beyond coding, Claude Opus 4.5 exhibits improvements in vision, mathematics, and research tasks. Through features like effort control on the Claude API, developers can now tailor performance output based on time and resource tradeoffs, while using significantly fewer tokens. Set to high effort, the model beats its predecessor’s scores using less than half the output data volume.
Security and safety are also central to the release. Opus 4.5 features improved resistance to prompt injection attacks and lower rates of “concerning behavior” across misalignment evaluations, positioning it as Anthropic’s safest release to date. The company cited advanced safety evaluations and ongoing research under its Societal Impacts and Economic Futures program aimed at monitoring broader changes driven by AI.
Enhanced integration across tools and platforms accompanies the model’s launch, including updates to Claude Code, Claude for Excel, Chrome, and the developer platform. Users can now conduct multi-agent workflows in the desktop app or coordinate extended research tasks using Claude memory and subagent teams.