Anthropic Launches Claude Opus 4.8 With Better Coding and Cheaper Fast Mode

Anthropic has released Claude Opus 4.8, its latest flagship AI model and a direct upgrade over Claude Opus 4.7.

The new model focuses on better coding, stronger reasoning, more reliable agentic work, and improved knowledge tasks. Anthropic says Opus 4.8 should feel more dependable during long, complex workflows, especially when users ask it to write code, inspect its own work, or coordinate multi-step tasks.

Claude Opus 4.8 is available now across Claude endpoints. Developers can access it through the Claude API using the model name claude-opus-4-8.

Anthropic keeps Opus 4.8 pricing unchanged

Anthropic is keeping standard Opus 4.8 pricing at the same level as Opus 4.7. The model costs $5 per million input tokens and $25 per million output tokens.

The bigger pricing change affects fast mode. Opus 4.8 fast mode now costs $10 per million input tokens and $50 per million output tokens. Anthropic says this makes fast mode three times cheaper than previous fast-mode models, while still allowing the model to run up to 2.5 times faster.

Opus 4.8 aims to fix agentic reliability problems

One of the biggest upgrades is reliability during agentic work. Anthropic says Opus 4.8 is around four times less likely than Opus 4.7 to ignore flaws in code it has written.

Anthropic is positioning Opus 4.8 as a more careful collaborator, not just a stronger benchmark model. The company says early testers found it better at flagging uncertainty, questioning weak plans, and avoiding unsupported claims.

Claude Code gets dynamic workflows

Alongside Opus 4.8, Anthropic introduced dynamic workflows in Claude Code. The feature is launching as a research preview and targets large engineering tasks that are too broad for a normal one-agent session.

With dynamic workflows, Claude can plan a large task and run hundreds of parallel subagents in one session. These subagents can inspect different parts of a codebase, test possible fixes, and verify results before Claude reports back to the user.

Dynamic workflows are available for Claude Code Enterprise, Claude Code Team, and Claude Code Max users.

Effort controls come to Claude

Anthropic is also adding effort controls to claude.ai and Cowork. Users can now choose how much effort Claude should spend on a response.

Opus 4.8 defaults to high effort, which Anthropic says offers the best balance between quality and user experience. Users can also choose extra or max effort when they want deeper reasoning, especially for difficult coding tasks or long-running workflows.

Lower effort settings should produce faster responses and use fewer rate-limit resources. Higher effort settings can improve answer quality, but they may consume more tokens.

API changes for developers

Claude Opus 4.8 also brings a developer-focused Messages API update. The API now supports system entries inside the messages array, allowing developers to update instructions mid-task without routing those changes through a user message.

This should help agent developers adjust permissions, token budgets, and environment context while a task runs.

Mythos-class models are coming soon

Anthropic also confirmed that Mythos-class models will come to all customers in the coming weeks. Mythos AI recently appeared in public Claude tools ahead of launch, suggesting Anthropic is preparing a broader rollout of its next higher-intelligence model class.

For Anthropic, Opus 4.8 looks like a targeted upgrade rather than a complete model overhaul. It keeps pricing steady, lowers fast-mode costs, improves agentic reliability, and gives Claude Code a more ambitious workflow system for large engineering projects.

In related AI coding news, Microsoft is reportedly reducing internal reliance on Claude Code and pushing more developers toward GitHub Copilot CLI.