Anthropic has released Claude Opus 4.8, introducing performance improvements, stronger reliability, and new user controls. The launch also brings dynamic workflows for Claude Code and configurable effort levels across Claude applications.
Anthropic has released Claude Opus 4.8, the latest version of its flagship AI model, alongside a set of new features aimed at developers, enterprises, and everyday users. The update builds on Claude Opus 4.7 with improvements across coding, reasoning, agentic workflows, and knowledge-based tasks while maintaining the same pricing structure.
According to Anthropic, Opus 4.8 delivers stronger performance across internal benchmarks and offers improved reliability when handling complex multi-step tasks. Early testers reported that the model is more consistent when performing autonomous work and better at identifying uncertainty rather than making unsupported claims.
One of the most notable improvements focuses on model honesty and self-evaluation. Anthropic said Opus 4.8 is around four times less likely than its predecessor to overlook flaws in code it has generated. The company’s alignment testing also found lower rates of deceptive or otherwise misaligned behavior compared with Opus 4.7.
The release introduces several new capabilities across Anthropic’s product ecosystem. Claude Code gains a research-preview feature called Dynamic Workflows, which allows Claude to plan and coordinate hundreds of parallel subagents within a single task. Anthropic said the system can now handle large-scale software projects, including codebase migrations involving hundreds of thousands of lines of code, while validating outputs before reporting results.
The company is also adding effort controls across Claude.ai and Cowork. Users can now choose how much computational effort Claude devotes to a task, balancing response quality, speed, and rate-limit consumption. Higher settings allow the model to spend more time reasoning before generating answers, while lower settings prioritize faster responses.
Anthropic also updated its Messages API, enabling developers to modify system instructions during a running workflow without disrupting prompt caching or requiring a new user interaction.
Building More Autonomous AI Systems
The introduction of Dynamic Workflows reflects a broader industry shift toward AI systems capable of coordinating multiple specialized agents to complete complex objectives. Rather than generating a single response, these systems can plan work, distribute subtasks, verify outputs, and combine results into a final deliverable.
The approach is particularly relevant for software engineering, where companies increasingly expect AI tools to handle larger projects rather than isolated coding tasks. By enabling hundreds of parallel subagents, Anthropic is positioning Claude Code to compete more directly with emerging autonomous development platforms and enterprise engineering tools.
The new effort controls also give users greater flexibility in managing tradeoffs between performance and cost, a growing concern as advanced reasoning models consume more computing resources.
Preparing for the Mythos Era
The Opus 4.8 release arrives as Anthropic continues to develop a new generation of AI systems under Project Glasswing. The company said a limited number of organizations are already using Claude Mythos Preview, an advanced model designed for cybersecurity applications.
Anthropic has previously stated that Mythos identified more than 10,000 high- and critical-severity software vulnerabilities through security research efforts. According to the company, models at that capability level require additional safeguards before broader deployment.
The announcement also comes just as Anthropic raised a $65 billion Series H funding round at a $965 billion valuation, surpassing OpenAI to become Silicon Valley’s most valuable AI company. The financing is expected to support continued model development, infrastructure expansion, and commercialization of advanced systems such as Mythos.
The launch of Opus 4.8 therefore serves both as an upgrade to Anthropic’s existing flagship model and as a stepping stone toward future Mythos-class systems. Anthropic said it expects to make those higher-capability models more widely available in the coming weeks as safety and security measures continue to mature.
OpenAI has launched the Rosalind Biodefense program and expanded access to its GPT-Rosalind model for trusted developers and government partners. The initiative aims to accelerate biodefense, pandemic preparedness,…
