Anthropic released its flagship model "Claude Opus 4.8" on May 28, 2026[1]. The new model improves on Opus 4.7 across benchmarks while keeping prices unchanged, and for users on claude.ai and Cowork it ships with a new "effort control" that lets them choose how much effort Claude puts into each response—available on every plan[1]. Anthropic also cut the price of fast mode to roughly a third of the previous generation and added a research-preview feature called "Dynamic Workflows" to Claude Code for large-scale parallel execution[1].
A New "Effort Control" Sits Next to the Model Selector
The update most directly visible to everyday users in this Opus 4.8 release is the new effort control[1]. A new control now sits next to the model selector in claude.ai and Cowork, letting users specify how much effort Claude should spend on any given response[1].
At higher effort settings, Claude thinks more frequently and more deeply in order to lift answer quality[1]. At lower effort settings, responses come back faster and consume rate limits more slowly[1]. The user gets to decide, on a per-question basis, whether they want speed or depth. Notably, the effort control is not a perk reserved for Pro or Max—Anthropic has opened it up to users on all plans[1].
High by Default, "Extra" for Hard Problems, "Max" at the Top
Opus 4.8 defaults to "high" effort[1]. Anthropic positions high as "the best overall balance of quality and user experience," noting that on coding tasks it uses roughly the same number of tokens as Opus 4.7's default while delivering better performance[1].
For more demanding work, users can pick "extra" (labeled "xhigh" in Claude Code) or "max," and Claude will spend more tokens to polish the result[1]. Anthropic recommends "extra" for difficult tasks and long-running asynchronous workflows[1]. Claude Code rate limits have also been raised to accommodate the higher token usage of upper effort levels, so users can match the level to the nature of the project[1].
Fast Mode: 2.5× Speed, One-Third the Price
Fast mode, aimed at situations where response speed matters most, also received a significant overhaul in the Opus 4.8 generation. According to Anthropic, Opus 4.8's fast mode runs at about 2.5× the speed of regular mode, and its price has been cut to one-third of fast mode on previous models[1].
Specifically, regular-mode pricing matches Opus 4.7 at $5 per million input tokens and $25 per million output tokens, while fast mode runs $10 per million input tokens and $50 per million output tokens[1]. Developers can invoke the model as claude-opus-4-8 via the API, making it easier to choose "speed-first or quality-first" at the endpoint level as well[1].
Hallucination Control and Improved "Honesty"
Another notable upgrade in Opus 4.8 is what Anthropic calls "honesty." AI models sometimes declare success on thin evidence; Opus 4.8 is reportedly more inclined to flag its own uncertainty and avoid claims it can't back up[1].
In Anthropic's internal evaluations, Opus 4.8 is about four times less likely than Opus 4.7 to let flaws in its own code pass unremarked[1]. The alignment assessment also reports record-high scores on traits such as "supporting user autonomy" and "acting in the user's best interest," with rates of undesirable behaviors—deception or willingness to cooperate with misuse—substantially lower than Opus 4.7[1]. Details are documented in the Claude Opus 4.8 System Card, published the same day[1].
"Dynamic Workflows" Lands in Claude Code
On the developer side, Claude Code gains "Dynamic Workflows" as a research preview[1]. Claude plans the work, spins up hundreds of parallel subagents within a single session, and verifies its own outputs before reporting back to the user[1].
Combined with Opus 4.8, Anthropic says this lets Claude Code carry out codebase-scale migrations spanning hundreds of thousands of lines, from kickoff to merge, with the existing test suite as the bar[1]. Dynamic Workflows is available on Claude Code's Enterprise, Team, and Max plans[1]. The Messages API also now accepts system entries inside the messages array, so developers can update permissions, token budgets, or environment context mid-run without busting the prompt cache or routing the update through a user turn[1].
Summary
Claude Opus 4.8 is a distinctive update not just for the lift in benchmarks and honesty but for opening the new effort control—letting users choose how deep Claude thinks per response—to every plan. The release also brings a sizable price cut and a 2.5× speedup for fast mode, the Dynamic Workflows feature in Claude Code, and system-entry support in the Messages API. Opus 4.8 is live as of today on claude.ai, Cowork, and the API, identified as claude-opus-4-8, at the same regular-mode price as Opus 4.7 ($5 / $25 per million input/output tokens). Anthropic has also signaled that lower-cost models matching Opus-class capability and the higher-capability "Mythos"-class model—currently in limited release via Project Glasswing—are next on the roadmap, so the model-update cadence looks set to stay short.
