Anthropic releases Claude Opus 4.8 with broad gains

Anthropic has released Claude Opus 4.8, the latest version of its flagship AI model, introducing updates aimed at software development, autonomous workflows and enterprise use cases while keeping standard pricing unchanged.

The new model replaces Claude Opus 4.7 and is available immediately through Claude applications and the Claude API. Alongside the model upgrade, Anthropic has introduced new controls that allow users to adjust the amount of computational effort devoted to individual tasks and expanded workflow capabilities in Claude Code.

Model updates

Anthropic said Opus 4.8 delivers improvements across coding, reasoning, agentic workflows and knowledge-based tasks.

The company stated that one of the main areas of progress is model reliability. Internal evaluations found the model is less likely to make unsupported claims and more likely to identify uncertainties or potential errors in its own work.

Anthropic said Opus 4.8 is around four times less likely than its predecessor to overlook flaws in code without flagging them to users.

The company also reported improvements in alignment assessments, including measures related to user autonomy, user interests and rates of misaligned behaviour such as deception or cooperation with misuse.

Workflow tools

A key addition accompanying the release is a feature called Dynamic Workflows, currently available as a research preview within Claude Code.

The feature allows Claude to break large projects into multiple tasks handled by parallel agents operating within a single session. Anthropic said the system can plan work, run hundreds of sub-agents and verify outputs before presenting results to users.

According to the company, the capability can be applied to software engineering projects involving codebase migrations across hundreds of thousands of lines of code.

Claude Code users also gain access to new effort settings. These controls allow users to choose how much reasoning and computation the model applies to a task.

Higher effort settings are intended for more complex problems and longer-running workflows. Lower effort settings prioritise speed and reduce token consumption.

Anthropic said Opus 4.8 defaults to a high-effort mode, which it considers the best balance between response quality and efficiency.

API changes

The company also announced updates to the Messages API.

Developers can now place system instructions directly within the messages array, allowing guidance and operational parameters to be adjusted during a task without interrupting prompt caching or requiring a new user message.

Anthropic said the change could help developers modify permissions, token budgets and environment settings while agents are running.

Customer testing

Several technology companies and enterprise users participated in early testing of Opus 4.8.

"On our long-running evals, Claude Opus 4.8's analysis was consistently higher quality than prior Opus models. It finished faster and produced richer, more information dense outputs. Overall, a noticeably better signal to noise ratio. The biggest differentiator was Opus 4.8's tendency to proactively flag issues with the inputs and outputs of an analysis, something other models routinely missed and left to the users to catch," said Michael Ran, Sr. Investment Associate, Bridgewater.

"Claude Opus 4.8 is the strongest computer-use and browser-agent model we've tested, scoring 84% on Online-Mind2Web, which is a meaningful jump over both Opus 4.7 and GPT-5.5. It stays reflective and on-task in the way our customers' agent workloads need to be reliable end-to-end," said Miguel Gonzalez, Tech Lead, Browserbase.

"Claude Opus 4.8 uses tools cleanly and follows instructions with the consistency our autonomous engineering workloads need to keep running unattended. It improves on Opus 4.6 and fixes the comment-verbosity and tool-calling issues we saw with Opus 4.7. This release from Anthropic translates directly into faster capability gains for engineers building on Devin," said Scott Wu, CEO, Cognition.

Pricing plans

Anthropic said pricing for regular Opus 4.8 usage remains unchanged from Opus 4.7 at USD $5 per million input tokens and USD $25 per million output tokens.

The company also introduced a fast mode for Opus 4.8. Anthropic said the mode operates at up to 2.5 times the speed of standard operation and is three times cheaper than comparable fast modes offered for previous models.

Fast mode pricing is set at USD $10 per million input tokens and USD $50 per million output tokens.

Anthropic also said it is continuing work on future AI models that offer Opus-level capabilities at lower cost and is developing a higher-capability class of systems under its Project Glasswing initiative.

The company said a limited number of organisations are already using a preview version of its next-generation model, Claude Mythos Preview, for cyber security applications. Anthropic expects broader availability once additional cyber security safeguards are completed.