Claude Opus 4.8 Is Out: Everything That Changed (May 2026)

Anthropic released Claude Opus 4.8 on May 28, 2026, and it immediately hit the top of Hacker News with over 1,600 points. The model is available today on claude.ai, the Claude API, and major cloud platforms — at the same price as before.

This article covers exactly what changed, what is new, and whether it is worth switching to if you are currently using Claude Opus 4.7 or another model.

What is Claude Opus 4.8?

Claude Opus 4.8 is Anthropic's newest and most capable AI model. It is an upgrade to Claude Opus 4.7, with improvements across coding, agentic tasks, reasoning, and long-session work.

The headline: on Anthropic's own Super-Agent benchmark, Opus 4.8 is the only model that completes every single case end-to-end. Not GPT-5.5. Not prior Opus versions. Only Opus 4.8 — at the same cost as GPT-5.5.

This is the version of Claude that is going to matter if you are building anything with AI agents, running Claude Code for software projects, or relying on Claude for complex multi-step work.

What is New in Claude Opus 4.8?

Anthropic shipped four significant updates alongside the model itself.

1. Effort Control on claude.ai

Users on claude.ai now have direct control over how much effort Claude applies to a task.

This sounds simple but it is actually very useful. If you want a quick answer, you can tell Claude to use less effort and it responds faster. If you are tackling a genuinely hard problem — a complex analysis, a large coding task, a multi-step research project — you can push Claude to apply maximum effort and it will work deeper and longer on the problem.

This gives beginners and power users both something they have been asking for: more transparency and control over how Claude works.

2. Dynamic Workflows in Claude Code

Claude Code now has a dynamic workflows feature. This allows Claude Code to handle very large-scale engineering problems by breaking them into adaptive sub-tasks and managing the execution flow automatically.

Early testers at Cursor found that Claude Opus 4.8 exceeded prior Opus models on CursorBench across every effort level, with more efficient tool calling — meaning Claude gets to the right result in fewer steps.

For developers using Claude Code to build or maintain large codebases, this is the biggest practical improvement in this release.

3. Fast Mode: 2.5× Speed, 3× Cheaper

Opus 4.8's fast mode lets the model work at 2.5 times its standard speed. What changed from the previous version: fast mode is now three times cheaper than it was on prior Opus models.

This makes Opus 4.8 the most accessible high-performance Claude version to date. If cost was previously a barrier to running Opus-level intelligence on a project, that barrier just got significantly lower.

4. Better Judgment in Agentic Tasks

Multiple early testers reported the same pattern independently: Opus 4.8 has noticeably better judgment when performing autonomous tasks.

Specifically, it:

Asks clarifying questions before making irreversible changes
Catches its own mistakes during a task and corrects them
Pushes back on plans that are not sound, rather than blindly executing
Builds up context and confidence before taking big actions

This matters a lot for anyone using Claude for automation. A model that checks its own work and flags issues is far more useful than one that plows ahead and makes expensive mistakes.

Claude Opus 4.8 Benchmark Results

Here is how Opus 4.8 performed on key benchmarks, with verified results from the Anthropic announcement:

Super-Agent Benchmark Claude Opus 4.8 is the only model to complete every case end-to-end. This benchmark tests full agentic execution across real-world workflows: translation, deep research, slide-building, and analysis. GPT-5.5 did not match it at comparable cost.

Online-Mind2Web (Browser Agent) Opus 4.8 scored 84% — the highest of any model tested, meaningfully above both Opus 4.7 and GPT-5.5. This benchmark tests how well AI agents navigate real websites to complete tasks.

CursorBench (Coding) Opus 4.8 exceeds prior Opus models at every effort level, with more efficient tool calling and fewer wasted steps.

Legal Agent Benchmark Opus 4.8 holds the highest recorded score and is the first model to break 10% on the all-pass standard (where every sub-step of a legal task must pass, not just the final result). For complex legal AI applications, that is a meaningful threshold.

Effort Levels Explained for Beginners

One of the most beginner-friendly additions in this release is effort control. Here is a simple breakdown:

Low effort — Great for quick questions, summarizations, and simple lookups. Responds faster and uses fewer compute resources.

Standard effort — The default. Good for most everyday tasks: writing emails, explaining concepts, reviewing documents.

High effort — For complex tasks where you want Claude to think deeply: multi-step analysis, research projects, difficult coding problems, strategic planning.

You set effort inside claude.ai directly. No special prompting required. This removes a common frustration where Claude would sometimes rush through a hard problem or spend too long on a simple one.

Who Should Care About Claude Opus 4.8?

Developers building with Claude Code: Dynamic workflows and better judgment make this a meaningful upgrade for large-scale engineering work. If you are maintaining or building a large codebase with Claude, upgrade.

AI automation builders: Smarter agentic behavior means fewer errors in automated workflows. The improvement in catch-and-correct behavior is especially valuable if Claude is running tasks without a human watching every step.

Content creators and researchers: Long-session context handling is better — Claude maintains style, voice, and direction more consistently across an extended conversation.

Budget-conscious users: Fast mode at 3× lower cost than before makes Opus 4.8 accessible for more use cases that previously felt expensive.

Beginners exploring Claude: Effort control makes the model more intuitive. You now have a simple dial for how hard Claude should work on something, without needing to engineer complex prompts.

Claude Opus 4.8 vs Claude Opus 4.7: Is the Upgrade Worth It?

If you are currently using Opus 4.7, the upgrade case looks like this:

Feature	Opus 4.7	Opus 4.8
Agentic task completion	Solid	Best-in-class (Super-Agent benchmark)
Fast mode cost	Higher	3× cheaper
Fast mode speed	2.5×	Same
Effort control	No	Yes
Dynamic workflows (Claude Code)	No	Yes
Browser agent (Mind2Web)	Good	84% (best tested)
Tool calling efficiency	Good	More efficient
Price	Same	Same

The price is unchanged. There is no cost reason to stay on 4.7. If you are in the API, update your model string to claude-opus-4-8. If you are on claude.ai, the model is already live.

How to Access Claude Opus 4.8

claude.ai users: The model is available now. If you have a Pro or Max subscription, you can select Opus 4.8 in the model picker. Effort control is available directly in the chat interface.

API users: Use model ID claude-opus-4-8-20260528 or check the Anthropic API documentation for the exact string. Pricing is the same as Opus 4.7.

AWS Bedrock and Google Vertex users: Opus 4.8 is available on both major cloud platforms as of today.

If you are new to the Claude API and want to get started, check out our guide: How to Use the Claude API for Beginners.

How Opus 4.8 Fits in the Claude Model Lineup

Anthropic has several Claude models available simultaneously, each for different needs:

Claude Haiku — Fastest and cheapest. Best for high-volume simple tasks. Claude Sonnet — Balanced. Strong everyday performance at moderate cost. Claude Opus 4.8 — Most capable. Best for complex, agentic, and long-running work.

For most beginners, Sonnet is the starting point. But if you are using Claude for automation, coding, or any serious research workflow, Opus is worth the cost — especially now that fast mode is 3× cheaper.

What This Means for AI in 2026

The race for best AI model in 2026 continues to move fast. GPT-5.5 from OpenAI has been the main benchmark to beat. Opus 4.8 does not just match it — it beats it on agent completion at the same cost.

Two trends stand out in this release:

Agentic AI is now serious. Anthropic, OpenAI, and Google all have agents as their primary development focus. The Super-Agent benchmark, browser agents scoring 84%, and dynamic workflows in Claude Code all point the same direction: AI that can complete real multi-step tasks reliably is getting close.

Speed and cost are converging. Fast mode at 2.5× speed for 3× less cost is not a minor footnote. It suggests the underlying infrastructure for running frontier models is getting significantly more efficient. That is good for everyone building with AI.

Frequently Asked Questions

What is Claude Opus 4.8?

Claude Opus 4.8 is Anthropic's newest and most capable AI model, released May 28, 2026. It improves on Opus 4.7 with better agentic performance, effort control, faster and cheaper fast mode, and dynamic workflows in Claude Code.

Is Claude Opus 4.8 free?

Claude Opus 4.8 is available on claude.ai with a Pro or Max subscription. API access requires an Anthropic API key and is billed by token. The pricing is the same as Opus 4.7.

What is the Super-Agent benchmark?

The Super-Agent benchmark tests whether an AI can complete real agentic workflows end-to-end without human intervention — tasks like research, slide-building, deep analysis, and translation pipelines. Opus 4.8 is the only model to pass every case.

How fast is Claude Opus 4.8 fast mode?

Fast mode runs at 2.5× the speed of standard Opus. With Opus 4.8, fast mode is also 3× cheaper than it was for prior Opus versions.

What is effort control in Claude Opus 4.8?

Effort control is a new feature in claude.ai that lets you set how much thinking and work Claude applies to a response. Low effort for quick answers, high effort for complex problems. No prompting tricks needed — it is a direct setting in the interface.

Is Claude Opus 4.8 better than GPT-5.5?

On Anthropic's Super-Agent benchmark, Opus 4.8 outperforms GPT-5.5 at the same cost. On Online-Mind2Web (browser agent tasks), Opus 4.8 scores 84% versus lower scores for GPT-5.5. On some other benchmarks, results are closer. For agentic and coding work, Opus 4.8 appears to have an edge as of this release.

Where is Claude Opus 4.8 available?

Claude Opus 4.8 is available on claude.ai, the Anthropic API, AWS Bedrock, and Google Vertex AI as of May 28, 2026.

Should I switch from Opus 4.7 to Opus 4.8?

Yes. The price is the same, and Opus 4.8 is better across every benchmark tested. There is no reason to stay on 4.7.