Anthropic Releases Claude Opus 4.8: Faster, More Honest, and Signalling Mythos Is Nearly Ready
Claude Opus 4.8 introduces Dynamic Workflows for Claude Code in research preview, a new effort control system for claude.ai and Cowork, and a Fast Mode that runs at 2.5x speed at three times lower cost, all at unchanged pricing.
Anthropic released Opus 4.8 on Thursday, May 28, 2026, the newest version of its most advanced publicly available model, arriving just 41 days after Opus 4.7, a much faster upgrade cycle than normal for Anthropic.
The accelerated release may reflect the lukewarm response to Opus 4.7, alongside mounting competition from GPT-5.5-powered Codex and Google’s Gemini Flash. As the Anthropic confirmed, pricing remains identical to Opus 4.7, $5 per million input tokens and $25 per million output tokens, removing migration friction for teams already using the Opus rate card.
The model is live immediately across the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
What Dynamic Workflows Change for Claude Code
The notable addition in Opus 4.8 is Dynamic Workflows, a research preview feature for the evolving Claude Code that enables codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge, using the existing test suite as its quality benchmark.
A single run can spawn up to 1,000 parallel subagents, with resumable runs meaning an interrupted workflow does not need to restart from scratch, a practical fix for long-running agentic coding jobs that previously had no recovery path if a session was interrupted.
In one confirmed test case, early testers used Dynamic Workflows to automatically rewrite and clean up massive amounts of code across multiple files all at once, the kind of task that previously required weeks of human engineering time.
Honesty Gains and Effort Control of Opus 4.8
Alongside Dynamic Workflows, Anthropic is launching effort control for claude.ai and Cowork, letting users choose how much reasoning Claude applies to a response, with high-effort settings thinking longer and low-effort settings using fewer tokens and slowing rate-limit usage.
The behavioural improvements in Opus 4.8 are as significant as the infrastructure ones.
Early testers report the model is around four times less likely than its predecessor to allow flaws in code it has written to pass without comment, a gain that directly addresses one of the most persistent frustrations developers report when using AI coding assistants at scale.
Additionally, an upgraded Fast Mode allows Opus 4.8 to work at 2.5x the speed while being three times cheaper than previous fast-mode iterations, making the flagship tier practically accessible for use cases that could not previously justify the compute cost.
What Mythos Means for What Comes Next
The release carries a signal beyond its own feature set. Anthropic hinted that the Mythos preview period might soon end: “We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks.”
Opus 4.8 still lags the performance of Mythos, but the gap between what enterprises can access today and what Anthropic has built is, by its own account, now measured in weeks rather than months.
For anyone tracking how Anthropic’s AI development has accelerated through 2026, Opus 4.8 is both a competitive response and a runway marker for what arrives next.
Source: Introducing Claude Opus 4.8
