Anthropic Launches Claude Opus 4.8 With Improved Coding and New Effort Controls

Anthropic has launched Claude Opus 4.8, bringing incremental improvements across coding, agentic tasks, reasoning, and practical knowledge work. The updated model builds on the foundation of its predecessor while introducing new user controls and maintaining the same standard pricing.



A central focus of the release is what Anthropic describes as increased honesty and sharper judgment during agentic tasks. Anthropic says Opus 4.8 is roughly four times less likely than Opus 4.7 to allow flaws in its own code to pass without comment. Early testers reported the AI is now more likely to flag uncertainties in its work rather than confidently making unsupported claims.


Anthropic also highlighted improvements in model alignment, saying internal evaluations found substantially lower rates of deceptive or misuse-related behavior compared to Opus 4.7, with performance approaching that of the company's Claude Mythos Preview model.

For developers handling large projects, Anthropic is previewing a dynamic workflows feature in Claude Code for Enterprise, Team, and Max subscribers. This allows the model to plan work and run hundreds of parallel subagents in a single session before verifying results and reporting back to the user. The company says the feature can handle codebase-scale migrations spanning hundreds of thousands of lines of code from kickoff to merge.

To give users more flexibility over how the AI operates, Anthropic added a new effort control dial on claude.ai and Cowork. Users can reduce effort for faster responses that consume rate limits more slowly or increase it to extra and max settings for more thorough reasoning on complex tasks. The feature is available across all plans. Fast mode for Opus 4.8, which operates at 2.5 times the speed of the standard model, is now three times cheaper than it was for previous models.

Developers using the Claude API are also getting a requested capability. The Messages API can now accept system entries directly inside the messages array. This lets instructions, token budgets, permissions, or environment context be updated during a task without breaking the prompt cache.


The update is available everywhere starting today. Regular usage remains priced at $5 per million input tokens and $25 per million output tokens. Anthropic also noted that fast mode pricing is $10 per million input tokens and $50 per million output tokens.

Looking ahead, Anthropic said it is making swift progress toward releasing a new class of models with capabilities beyond Opus. As part of Project Glasswing, Claude Mythos Preview remains available to a small number of organizations performing cybersecurity work. Anthropic says Mythos-class models require stronger cybersecurity safeguards before a broader release and expects those models to become available to all customers in the coming weeks.