Claude Opus 4.8: What Changed, What It Costs, and What It Means for Builders

Anthropic released Claude Opus 4.8 on May 28, 2026. It builds on Opus 4.7, ships at the same price, and is available everywhere on day one through claude.ai, the API (model name claude-opus-4-8), and the major cloud platforms. [1]

Anthropic describes it as “a modest but tangible improvement,” which is a more honest summary than most launch posts get. [1] The headline is not a giant capability leap. It is a better model at the same price, plus a set of controls that change how much you spend to use it.

For anyone building on these models, the changes that matter most are not on the benchmark table. They are in what it costs to run. Here is what shipped and what it means.

Pricing stays flat

Regular usage is unchanged from Opus 4.7: $5 per million input tokens and $25 per million output tokens. Fast mode, where the model runs at about 2.5 times the speed, now costs $10 per million input and $50 per million output, which Anthropic says is three times cheaper than fast mode was on previous models. [1] Anthropic also lists up to 90 percent savings with prompt caching and 50 percent with batch processing. [3]

For anyone running real volume, a faster tier at a third of the old fast-mode cost is the most concrete thing in this release.

A new effort control

Claude now has an effort setting that decides how hard it works on a given response. It is available on all plans in claude.ai and Cowork. [1]

• High is the default and Anthropic’s recommended balance.
• Extra (called xhigh in Claude Code) and Max spend more tokens for better results on hard or long-running tasks.
• Lower settings answer faster and use your rate limits more slowly.

The practical read: you can tune cost against quality per task instead of paying the same for a quick lookup and a deep analysis. That is the kind of lever that matters when AI is a line item in your budget, not a demo.

Better coding, and a focus on “honesty”

On capability, Anthropic reports gains in coding, reasoning, and agentic work. Its SWE-Bench Pro coding score rises to 69.2 percent from 64.3, and its GDPval knowledge-work score climbs to 1890 from 1753. [1][7]

The trait Anthropic leads with is honesty. The company says Opus 4.8 is more likely to flag uncertainty and, by its own testing, about four times less likely than 4.7 to let a flaw in its own code slip by unremarked. [1][6] These are Anthropic’s internal results, so read them as a strong signal rather than an independent verdict. The direction is the right one, though. A model that admits when it is unsure is more useful than one that sounds confident and is wrong.

Two quieter updates for developers

Two changes will matter more to people building on the API than to casual users.

• Dynamic workflows in Claude Code, a research preview on Enterprise, Team, and Max plans, let Claude plan a large job, run hundreds of subagents in parallel, and verify its own output before reporting back. It is aimed at large codebase migrations. [1][2]
• The Messages API now accepts system instructions inside the messages array, so you can update Claude’s instructions mid-task without breaking the prompt cache. If you design the context that agents run on, that is a genuinely useful primitive. [1]

What about Mythos?

Opus 4.8 is not Mythos. Mythos is Anthropic’s more powerful, cybersecurity-focused model, still limited to a small group of partners through Project Glasswing. Anthropic says it expects to bring Mythos-class models to all customers within the coming weeks, once stronger safeguards are in place. [1] For now, Opus 4.8 is the most capable model you can actually use.

How the coverage is landing

A few hours in, most outlets are reading this the same way. Axios framed the launch as less about raw power than about giving customers ways to tailor what they spend. [4] TechCrunch noted the unusually quick turnaround, 41 days after a 4.7 release that drew a lukewarm response. [5] Not everyone is sold on the headline trait: Techzine pushed back on the honesty framing, pointing out that it mainly means the model’s words line up better with what it actually did, and that Opus 4.8 is still far from the capability Anthropic attributes to Mythos. [8] Early developer reaction on Hacker News was more muted, with commenters describing the recent run of point releases as modest and hard to feel in day-to-day use. [9] The independent practitioners who usually publish hands-on reviews had not weighed in at the time of writing.

The bottom line

Claude Opus 4.8 is a sensible, incremental upgrade priced exactly like the one it replaces. If you chat with it occasionally, you will notice it is a little sharper. If you build with it, the effort control and the cheaper fast mode are the reasons to pay attention, because they change the math on running AI at scale.

---

FAQ

When was Claude Opus 4.8 released? May 28, 2026.

How much does Claude Opus 4.8 cost? Regular usage is $5 per million input tokens and $25 per million output tokens. Fast mode is $10 and $50 respectively.

What is new in Claude Opus 4.8 compared to Opus 4.7? Better coding, reasoning, and agentic performance, a new effort control, a faster and cheaper fast mode, improved honesty, and dynamic workflows in Claude Code. Anthropic calls it a modest but tangible upgrade.

What is the Claude Opus 4.8 API model name? claude-opus-4-8.

Is Claude Opus 4.8 the same as Claude Mythos? No. Mythos is a separate, more powerful model still restricted to Project Glasswing partners. Anthropic says Mythos-class models are coming in the weeks ahead.

---

References

[1] Anthropic, “Introducing Claude Opus 4.8,” May 28, 2026. https://www.anthropic.com/news/claude-opus-4-8

[2] Anthropic, “Introducing dynamic workflows in Claude Code,” May 28, 2026. https://claude.com/blog/introducing-dynamic-workflows-in-claude-code

[3] Anthropic, Claude Opus model and pricing overview. https://www.anthropic.com/claude/opus

[4] Axios, “Anthropic releases new model, Opus 4.8,” May 28, 2026. https://www.axios.com/2026/05/28/anthropic-opus-release-mythos

[5] TechCrunch, “Anthropic releases Opus 4.8 with new ‘dynamic workflow’ tool,” May 28, 2026. https://techcrunch.com/2026/05/28/anthropic-releases-opus-4-8-with-new-dynamic-workflow-tool/

[6] Inc., “Anthropic Says Its Claude Opus 4.8 Model Is Its ‘Most Honest’ Yet,” May 28, 2026. https://www.inc.com/ben-sherry/anthropic-says-its-latest-claude-model-is-the-most-honest-yet/91351657

[7] 9to5Mac, “Anthropic upgrades Claude with new Opus 4.8 model, here’s what’s new,” May 28, 2026. https://9to5mac.com/2026/05/28/anthropic-upgrades-claude-with-new-opus-4-8-model-heres-whats-new/

[8] Techzine, “Anthropic releases Claude Opus 4.8, promising a more honest model,” May 28, 2026. https://www.techzine.eu/news/applications/141667/anthropic-releases-claude-opus-4-8-promising-a-more-honest-model/

[9] “Claude Opus 4.8,” Hacker News discussion, May 28, 2026. https://news.ycombinator.com/item?id=48311647