Anthropic Unveils Claude Opus 4.8 and New Tools
Anthropic has released Claude Opus 4.8, the latest version of its flagship AI model, available immediately at the same standard price as its predecessor. The update brings benchmark improvements across coding, reasoning, and agentic tasks — and arrives alongside three platform features that expand how users and developers interact with the model.
Standard pricing stays at $5 per million input tokens and $25 per million output tokens. Fast mode, which runs the model at 2.5 times the standard speed, now costs $10 per million input and $50 per million output — a three-times price cut from what Opus 4.7 charged at that tier.
What's Different About Opus 4.8's Performance
Honesty is one of the sharpest improvements in this release. According to Anthropic, Opus 4.8 is roughly four times less likely than Opus 4.7 to let code flaws pass without flagging them — a direct fix for the tendency AI models have to project false confidence about their own work.
The company's alignment team assessed the model as reaching new highs on prosocial behavior, including supporting user autonomy and acting in the user's interest. Rates of misaligned behavior — such as deception or assisting with misuse — are described as substantially lower than Opus 4.7 and comparable to Claude Mythos Preview, Anthropic's most safety-optimized model to date. Full evaluation data is available in the Claude Opus 4.8 System Card.
Dynamic Workflows, Effort Control, and API Updates
Three new features ship alongside the model. Dynamic workflows, now in research preview inside Claude Code, lets Claude plan a task, spin up hundreds of parallel subagents in a single session, and verify outputs before returning results. Anthropic says this makes codebase-scale migrations possible across hundreds of thousands of lines of code — from kickoff to merge — using existing test suites as the quality bar. The feature is available on Enterprise, Team, and Max plans.
Effort control gives claude.ai and Cowork users a dial for how much computational work Claude applies to a response. Higher settings produce more thorough answers at the cost of rate limits; lower settings trade depth for speed. The control sits alongside the model selector and is available on all plans.
Developers also get a structural change to the Messages API. System instructions can now be passed inside the messages array rather than only at the top level, allowing mid-task permission or context updates without breaking prompt cache or routing changes through a user turn.
What Anthropic Says Is Coming Next
Anthropic describes Opus 4.8 as a tangible but incremental step. The company says it is working on models that match Opus-level capability at lower cost.
A separate, higher-intelligence model class is also in the pipeline. Under Project Glasswing, a limited group of organizations is already testing Claude Mythos Preview for cybersecurity work. Anthropic says stronger cyber safeguards are required before a wider release, and expects Mythos-class models to reach general availability within weeks.
Opus 4.8 is accessible now via the Claude API using the model string claude-opus-4-8. Read the full announcement on the Anthropic blog.