aiAI타임스 (AI Times)· 7/25/2026, 12:24:09 AM8.0

Anthropic Launches 'Claude Opus 5'...

Anthropic has launched 'Claude Opus 5,' a model that delivers top-tier intelligence at significantly reduced costs. The model outperforms 'Palm 5' in coding and agent benchmarks at half the price, shifting the Frontier models competition toward cost efficiency and everyday enterprise workflows. Opus 5 will be available to 'Claude Max' and 'Claude Pro' subscribers starting 24th, with access via cloud platforms including Amazon Bedrock, Google Cloud, and Microsoft Azure. Key features include near-Palm 5 performance for most enterprise tasks while maximizing cost efficiency, with high performance in coding and knowledge work benchmarks but slightly lower performance in cybersecurity tasks compared to 'Mistral 5.' The model operates more efficiently for routine tasks, maintaining pricing parity with the previous generation Opus 4.8 while offering half the cost of Palm 5. The cost-efficiency gains enable enterprise customers to achieve substantial cost savings for equivalent workloads. With a 1 million token context window, it can process hundreds to thousands of pages of documents in one go, and supports up to 128,000 output tokens. The 'Thinking' function, which enables internal reasoning processes, is enabled by default. The model also supports a 5-tier 'Effort Ladder' for precise control over intelligence levels and token consumption based on task complexity. A 'Fast Mode' option, operating approximately 2.5 times faster than default, addresses high-performance agent tasks. Additional utility features for agent developers include 'In-Context Tool Switching,' allowing tool additions or removals during conversations without compromising cache integrity, significantly reducing token costs and latency.

💡 AI analysis: Anthropic's launch shifts the LLM competition paradigm from breaking intelligence frontiers to cost-efficient enterprise workflow integration, accelerating the migration of enterprise users away from incumbent high-cost frontier models.
Related entities
View original (AI타임스 (AI Times)) →