Anthropic has released its new flagship artificial intelligence model Claude Opus 4.8, marking a new phase in the competitive race among leading AI developers to create agentic systems and corporate AI tools. The launch comes amid ongoing rivalry between Anthropic and competitors such as OpenAI and Google — all three companies are accelerating efforts to bring more powerful autonomous AI systems to the commercial market.
According to Anthropic, Claude Opus 4.8 is an updated version of the previous flagship Opus 4.7 model and demonstrates improved performance on numerous synthetic tests focused on agentic capabilities. These include autonomous programming, financial analysis, and computer operation tasks requiring multi-step reasoning and tool use. The company claims the model outperforms competing systems, including OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro, across several evaluation dimensions.
The key emphasis of the new version is on enhanced agentic behavior — the ability of AI systems to perform tasks with minimal human intervention. In practice, agentic AI serves as a semi-autonomous digital assistant capable of planning, executing, and adjusting complex workflows on behalf of the user. This capability is considered a primary driver of mass AI adoption in the corporate environment, where companies are exploring ways to automate research, software development, financial modeling, and administrative processes using AI agents.
Anthropic also emphasized improved model reliability, calling Opus 4.8 more 'honest' compared to previous versions. According to the company, the model better signals its own uncertainty, makes fewer unfounded claims, and more readily acknowledges a lack of evidence. This approach aims to address the pressing issue of large language model hallucinations — situations where AI generates plausible but incorrect information.
In addition to performance improvements, Anthropic introduced new system capabilities across its entire product ecosystem. One of these is dynamic workflows, a feature allowing users to launch numerous small helper agents working in parallel to solve complex tasks. The company notes that this enables more scalable automation, especially in corporate environments where workloads can be divided into distributed subtasks.
Another important update is adjustable 'compute cost level' parameters, allowing users to control the amount of computational resources allocated by the model to a specific task. A high cost level provides deeper reasoning and detailed results, while a low level provides speed and cost-efficiency. This directly affects token consumption, which determines the pricing of AI system usage.
Tokens are the basic units of measurement for AI input and output data, representing fragments of text or other media processed by the model. Anthropic notes that users can now manage costs more effectively by adjusting the processing intensity of Claude requests, making the system flexible for both simple short queries and highly complex tasks.
In terms of pricing, Anthropic has kept prices unchanged compared to the previous model. The rate is $5 per million input tokens and $25 per million output tokens; the accelerated mode with higher cost is priced at $10 per million input tokens and $50 per million output tokens. Anthropic confirmed immediate availability of the model, strengthening its strategy for scaling adoption among developers and corporate clients.
Dario Amodei, co-founder and CEO of the company, has repeatedly emphasized the company's focus on creating safer and more controllable AI systems, and the new product reflects this direction through an emphasis on transparency, reliability, and agent-based task execution.
Amid intensifying competition in the artificial intelligence industry, Claude Opus 4.8 becomes another step toward fully autonomous systems capable of solving cross-industry tasks, and also demonstrates the growing importance of efficiency, cost control, and reliability in commercial AI deployment.