Anthropic has just launched Claude Opus 4.8, the newest flagship in the Claude family. The release landed on May 28, 2026, and it brings something the company has been pushing for a while: an AI that is genuinely more honest about what it knows, alongside real gains in agentic work like coding, analysis, and computer use.
Opus 4.8 also introduces two new controls that change how you work with Claude: dynamic workflows, which let it run many small helpers in parallel, and effort controls, which let you dial up or down how hard the model thinks on a given task.
This guide explains Claude Opus 4.8 in plain language. What is new, how it compares to GPT-5.5 and Gemini 3.1 Pro, what it costs, and whether it is worth using right now. There is also a brief look at Claude Mythos, the second model Anthropic announced the same day. No hype, just the useful parts.
Here is the short version. Claude Opus 4.8 is Anthropic’s new top model, launched May 28, 2026. It is more honest, better at agentic tasks, and adds dynamic workflows and effort controls. Pricing stays the same as Opus 4.7, at $5 per million input tokens and $25 per million output tokens.
1. What Is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic’s newest top-tier AI model, released on May 28, 2026. The Yahoo Finance announcement confirms it sits at the top of the Claude line-up, replacing Opus 4.7 as the company’s most capable model.
In simple terms, you can think of Opus 4.8 as the model Anthropic uses for its hardest work. Tasks that involve careful reasoning, long pieces of code, complex analysis, or letting an AI agent run for a while on its own. The Sonnet and Haiku models in the family stay around for cheaper or faster jobs, but Opus is where Anthropic puts its newest research.
You can use it through Claude.ai, the Anthropic API, Claude Cowork, Claude in your browser, Claude Design, and partners like Cursor and GitHub. If you already have a paid Claude subscription, you may already see Opus 4.8 in your model picker.
2. What Is New in Opus 4.8: The Key Upgrades
Four changes do most of the work in this release. Each one is a meaningful shift, not just a number bump.
| Upgrade | What It Does | Why It Matters |
| Honesty | Flags uncertainty and avoids unsupported claims | Less fabrication, more reliable answers |
| Agentic skills | Stronger at autonomous coding, analysis, and computer use | Better for tasks where Claude works on its own |
| Dynamic workflows | Runs many small subagents in parallel | Tackles big jobs by splitting them up |
| Effort controls | Tunes how hard the model thinks per task | Saves money and time when you do not need depth |
The next sections walk through each one in plain language.
3. The Honesty Upgrade: Why It Stands Out
The biggest single improvement in Opus 4.8 is something that does not show up on a benchmark chart. Anthropic says the new model is noticeably more honest than its predecessors.
In practice, that means two things. First, when Opus 4.8 is unsure about something, it tells you instead of guessing confidently. Second, it makes fewer unsupported claims, the kind of statement that sounds right but cannot really be backed up.
Anthropic has been blunt about this. The company acknowledged that AI models often jump to conclusions and confidently claim progress on thin evidence. Opus 4.8 is trained to push back against that habit and to call out its own uncertainty when it has one.
Why this matters to you: If you use Claude for research, code review, or any high-stakes work, fewer confidently wrong answers is a real upgrade. It is not perfect, and you still need to verify important outputs, but the model is now actively trying not to bluff.
4. Dynamic Workflows and Effort Controls Explained
Two new capabilities change how you can use Claude day to day.
Dynamic Workflows
Anthropic now lets Claude set up and run hundreds of small subagents on a single task. Each subagent gets a piece of the work, runs in parallel, and feeds results back to the main agent for the final answer.
Imagine asking Claude to review a long codebase for security issues. Instead of one big sweep, it spins up many smaller workers, one per file or area, and combines what they find. The result is faster, more thorough, and cleaner to read.
Effort Controls
The second new control lets you tell Claude how much effort to put into a task. More effort means the model thinks longer and goes deeper. Less effort means it answers quickly and uses fewer tokens, which saves money.
This works inside Claude in your browser and inside Claude Cowork. So you can give a careful design review a higher effort setting, and a quick summary a lower one. You spend more only where it pays off.
These two features are clearly tied to the wider rise of agentic AI, where models do real work over time instead of just answering single questions.
5. Performance: How Opus 4.8 Compares to GPT-5.5 and Gemini 3.1 Pro
Anthropic says Opus 4.8 tops OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro on a number of synthetic AI benchmarks. The lead is strongest in three areas:
- Agentic coding. Letting Claude write, run, and fix code on its own across many steps.
- Agentic financial analysis. Working through long financial documents and producing usable analysis with limited supervision.
- Agentic computer use. Operating a computer through clicks, typing, and apps to complete real tasks.
Honest note: These are benchmarks the lab itself ran, which is normal in the industry but worth knowing. Independent tests from researchers and developers usually arrive within a few weeks. Treat the lead as real but provisional until then.
6. Pricing and How to Access Opus 4.8
Anthropic kept the price the same as Opus 4.7, which is a quiet win. You get a stronger model for the same cost per token.
| Plan | Input (per million tokens) | Output (per million tokens) |
| Standard | $5 | $25 |
| Fast mode | $10 | $50 |
Fast mode is for cases where speed matters more than cost. Standard mode is the right choice for most work.
If you only want to try Opus 4.8, the easiest path is the model picker inside Claude.ai. If you build with it, the same endpoints in the Claude API now serve Opus 4.8 when you select it.
7. What About Claude Mythos?
Anthropic also confirmed, in coverage carried by Reuters via KSL, that another model called Claude Mythos will roll out to all customers in the coming weeks.
Mythos is a separate large language model focused on advanced cybersecurity work. For now, it is being used inside something Anthropic calls Project Glasswing, with companies like Amazon, Microsoft, and Apple using it for security tasks under a controlled programme.
The model has also raised concerns among executives and world leaders about its capabilities, which is why the staged release matters. For most readers, Opus 4.8 is the day-to-day model. Mythos is the one to watch as the rollout opens up.
8. Should You Use Opus 4.8 Right Now?
Whether to switch depends on how you use Claude.
If you use Claude for everyday writing or simple coding
Switching is low risk. Same price, more honest answers, better quality on harder questions. There is almost no reason not to use it for your main work.
If you build with the Claude API or use Claude Cowork
This is where Opus 4.8 shines. The agentic improvements, dynamic workflows, and effort controls are aimed straight at builders. Expect noticeable gains on multi-step tasks, code reviews, and long autonomous runs.
If you mostly used Claude through Claude Design
The same upgrade applies, since Claude Design now runs on the latest Opus generation behind the scenes. Expect more accurate first drafts and better adherence to your brand instructions.
Catches to know
- Fast mode is more expensive, so use it only when speed truly matters.
- Benchmarks are from Anthropic for now. Independent reviews are coming.
- Effort controls only work in Claude in browser and Claude Cowork at launch.
- Honesty improvements are real but not magic. You still need to verify important outputs.
Frequently Asked Questions
What is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic’s latest flagship AI model, released on May 28, 2026. It is an upgrade over Opus 4.7 with improvements in honesty, agentic capabilities, and new features like dynamic workflows and effort controls.
When was Claude Opus 4.8 released?
Anthropic released it on Thursday, May 28, 2026, alongside an announcement about the upcoming Claude Mythos model.
How much does Claude Opus 4.8 cost?
It costs the same as Opus 4.7: $5 per million input tokens and $25 per million output tokens for the standard plan. Fast mode costs $10 per million input tokens and $50 per million output tokens.
How is Opus 4.8 different from Opus 4.7?
It is more honest, since it flags uncertainty and avoids unsupported claims. It is stronger on agentic tasks like coding, financial analysis, and computer use. And it adds two new controls: dynamic workflows for parallel subagents, and effort controls to tune how hard Claude thinks per task.
Does Opus 4.8 beat GPT-5.5 and Gemini 3.1 Pro?
Anthropic claims it tops both on a number of synthetic AI benchmarks, especially in agentic coding, agentic financial analysis, and agentic computer use. Independent tests are still to come, so treat the lead as real but early.
What is Claude Mythos?
Mythos is a separate Anthropic model focused on advanced cybersecurity. It is currently used inside Project Glasswing by Amazon, Microsoft, and Apple. Anthropic plans to roll it out to all customers in the coming weeks.
How do I try Claude Opus 4.8?
Open Claude.ai and choose Opus 4.8 from the model picker if you have a paid plan. Builders can select it through the Claude API. It also powers the latest versions of Claude Cowork, Claude in your browser, and Claude Design.
