Anthropic Launches Claude Opus 4.8 with Enhanced Capabilities
Anthropic has officially launched Claude Opus 4.8, the latest upgrade to its flagship AI model, promising notable improvements in coding, reasoning, reliability, and collaboration.
Building on Claude Opus 4.7, the company says the new model is more effective at tackling complex tasks while maintaining the same price as its predecessor.
The launch of Opus 4.8 comes alongside a range of new features for developers and everyday users, including effort control settings on Claude.ai, faster model performance, and dynamic workflows in Claude Code for handling large-scale software projects.
Performance Benchmarks
According to Anthropic, Claude Opus 4.8 delivers measurable gains across several major benchmarks. In agentic coding, the model achieved 69.2%, while its performance on Terminal Bench 2.1 for terminal-based coding tasks reached 74.6%.
The model also demonstrated stronger reasoning capabilities. On Humanity's Last Exam, a multidisciplinary reasoning benchmark, Claude Opus 4.8 scored 49.8% without tools and 57.9% with tools enabled.
In OS World Verified, which measures agentic computer use, the model recorded 83.4%. Meanwhile, its score in GDPval-AA for knowledge work stood at 1890, while Finance Agent v2 testing showed a 53.9% performance in agentic financial analysis.
Improved Reliability and Alignment
Anthropic says early testers found that Claude Opus 4.8 was more dependable and sharper in judgment, particularly during autonomous or agentic tasks. The company highlighted that the model is more likely to acknowledge uncertainty and significantly less likely to make unsupported claims.
In internal evaluations, Anthropic found that Opus 4.8 was nearly four times less likely than Claude Opus 4.7 to overlook flaws in code it generated. The company also stated that its alignment team observed stronger prosocial behaviour, with the model better supporting user autonomy and acting in users' best interests.
Safety Testing
Safety and alignment remain a key focus for Anthropic. Before launch, the company conducted extensive safety testing and said Claude Opus 4.8 showed substantially lower rates of harmful or misaligned behaviour, including deception and misuse assistance, compared to Opus 4.7.
Anthropic added that the model's alignment performance is comparable to Claude Mythos Preview, an advanced frontier model currently limited to select trusted partners under Project Glasswing.
New Features
Dynamic Workflows
One of the biggest additions arriving with Opus 4.8 is Dynamic Workflows, a research preview feature available in Claude Code. The tool enables Claude to break down massive software engineering challenges by planning tasks and deploying hundreds of parallel sub-agents within a single session.
Anthropic says this could allow developers to carry out large-scale codebase migrations involving hundreds of thousands of lines of code, from project kickoff to final merge, while using an existing test suite as a benchmark for success. The feature is currently available to Claude Code users on Enterprise, Team, and Max plans.
Effort Control
The company is also introducing Effort Control on Claude.ai and Claude Work. A new option alongside the model selector allows users to decide how much effort the AI should put into generating responses. Higher effort settings enable deeper thinking and more detailed outputs, while lower effort modes prioritise faster replies and lower rate-limit usage.
API Updates
Developers are also getting more flexibility through an update to the Messages API, which now supports system entries within message arrays. This enables mid-task instruction updates, such as permission changes, token budgets, or environment context, without interrupting prompt caching or requiring a user interaction.
Fast Mode
Anthropic also announced a new Fast Mode for Claude Opus 4.8, enabling the model to run 2.5 times faster. The company claims the mode is now three times cheaper than comparable options offered with earlier generations.
Pricing and Availability
Claude Opus 4.8 is available globally starting today. Pricing remains unchanged from Opus 4.7 for standard usage at $5 per million input tokens and $25 per million output tokens.
Fast Mode pricing is set at $10 per million input tokens and $50 per million output tokens. Developers can access the model through the Claude API using "claude-opus-4-8".
While Anthropic's benchmark scores and internal evaluations paint an optimistic picture of Claude Opus 4.8, independent third-party testing will likely offer a clearer assessment of how meaningful the improvements are in real-world usage.
Still, the launch marks another step forward in the rapidly evolving race among AI companies to build smarter, faster, and more reliable models.
