Do you use Claude every day? Good news: it has just got even better.

On 28 May 2026, Anthropic launched Claude Opus 4.8 - and this update really makes a difference. A new effort control, parallel agents, stronger honesty… with no change in price.

‍

Here is everything it changes for your teams 👇

‍



What really changes with Claude Opus 4.8


Opus 4.8 was released on 28 May 2026 - at the same price as before

First piece of good news: Claude Opus 4.8 is available everywhere from today, at the same price as its predecessor  $5 per million input tokens, $25 for output.

Same budget. Better model. That is rare, so it is worth noting.
‍

Try Claude Opus 4.8

‍

It is 4× more honest, and that is the real revolution

‍

raw Alignment opus48 · HTML Claude Opus 4.8 — Alignment
Safety & alignment
Misaligned behaviour
Score out of 10 — lower is better • Hover for details
Highly aligned → ← Less aligned
Sonnet 4.6 2.58 / 10
Sonnet 4.6 is the most accessible model — its rate of misaligned behaviour remains the highest of the four.
Opus 4.7 2.48 / 10
Opus 4.7, the direct predecessor of Opus 4.8, has a score similar to Sonnet 4.6.
Opus 4.8 1.83 / 10
Opus 4.8 makes a significant leap — almost at the level of Mythos Preview, Anthropic's most advanced AI.
Mythos Preview 1.77 / 10
Mythos Preview — Anthropic's most powerful AI, available only to Project Glasswing. Soon available to everyone.
Key takeaway: Opus 4.8 almost reaches the alignment level of Mythos Preview — Anthropic's most advanced AI, not yet available to the public. A huge leap compared with Opus 4.7.
Source: Anthropic System Card, May 2026


This is the most fundamental change in this version.

For a long time, AI models had a major flaw: pressing ahead confidently even when they were unsure. Claude 4.8 fixes that.

Opus 4.8 is around 4 times less likely than its predecessor to let a flaw in the code slip through without flagging it.

In a business setting, that changes everything:

✔️ Fewer silent bugs in your deliverables

✔️ Claude tells you when it is not sure - like a real colleague

✔️ Less time spent manually checking its outputs

The first testers summed it up well: Opus 4.8 is more likely to flag uncertainty about its work and less likely to make unsupported claims.

‍

The effort control: you decide how much Claude "thinks"

‍

This is the most visible feature for everyday users.

A new control next to the model selector in claude.ai and Cowork lets you choose how much effort Claude puts into a response. On higher settings, Claude thinks more often and more deeply. On lower settings, Claude responds faster and uses fewer credits.

In practice, for your teams:

📧 Quick email → low effort, answer in seconds

📊 Strategic analysis or complex contract → maximum effort, in-depth reasoning

You stay in control. You choose the depth to suit the task.

‍

raw Try button opus48 · HTML Try Claude Opus 4.8

‍

2.5× faster - 3× cheaper in fast mode

Opus 4.8's fast mode lets the model work at 2.5 times normal speed, and it is now 3 times cheaper than on previous models.

For teams that use Claude at volume - customer service, content production, data analysis - that is an immediate gain.

"Dynamic Workflows": hundreds of agents in parallel

‍

This is the feature that will change life for technical teams and IT departments.

Claude can now plan a piece of work and launch hundreds of sub-agents in parallel in a single session, check its own results, then deliver the whole thing. Claude Code with Opus 4.8 can therefore carry out migrations across an entire codebase, hundreds of thousands of lines  from kick-off to merge.

What used to take weeks can now be done in a few hours.

‍

The benchmarks that hurt the competition

‍

‍

raw Benchmark opus48 · HTML Claude Opus 4.8 — Benchmarks
Performance — June 2026
Opus 4.8 vs the competition
Hover over a row to see details • Higher score = better
Benchmark
Opus 4.8
Opus 4.7
GPT-5.5
Gemini 3.1 Pro
Agentic coding
SWE-Bench Pro
69.2%
64.3%
58.6%
54.2%
Terminal (code)
Terminal-Bench 2.1
74.6%
66.1%
78.2%
70.3%
Multi-discipline reasoning
Humanity's Last Exam
57.9%
54.7%
52.2%
51.4%
Computer use
OSWorld-Verified
83.4%
82.8%
78.7%
76.2%
Knowledge work
GDPval-AA
1,890
1,753
1,769
1,314
Financial analysis
Finance Agent v2
53.9%
51.5%
51.8%
43.0%
Best score
Exception: GPT-5.5 ahead on Terminal
Other models
Source: Anthropic, May 2026 — independent public benchmarks

‍

‍

What it changes for your teams, by role

💼 Finance & Analysis: Denser multi-document analysis, more reliable outputs, fewer unflagged errors in figures.

⚖️ Legal: Better consistency over long workflows, greater precision when reading contracts and clauses.

📣 Marketing & Communications: Claude keeps your style and tone better over long sessions. Smoother content production.

💻 IT & Dev: Large-scale code migrations, automated workflows, bugs flagged before they go unnoticed.

‍

Opus 4.8 is already impressive. But Anthropic is announcing something bigger.

Anthropic plans to launch a new class of model with intelligence even beyond Opus: Claude Mythos. A small number of organisations are currently using it for cybersecurity work as part of Project Glasswing. Models at this level need stronger cyber safeguards before being released to the general public. Anthropic plans to make them available to all its customers in the coming weeks.

Stay tuned - we will publish a full article as soon as Mythos is available.


‍

Claude Opus 4.8 is available everywhere today via claude.ai, Claude Code and the Claude API with the identifier claude-opus-4-8.

The AI Sisters verdict

2026 is the year AI moves from experimentation to industrialisation. Claude Opus 4.8 is not a technological curiosity; it is a more reliable, faster, more honest colleague. Available now, for the same price.

Want to roll out Claude across your teams? Our corporate Claude training starts from your role-based use cases.

To see where this version fits among Anthropic's other announcements, read all the new Claude features of 2026.

raw Faq opus48 · HTML FAQ — Claude Opus 4.8
FAQ
Everything you need to know about Claude Opus 4.8
Answers to the questions companies are asking about Anthropic's new model.
What is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic's most advanced AI model, released on 28 May 2026. It succeeds Opus 4.7 and brings better performance on coding, agentic tasks, reasoning and knowledge work — at the same price as the previous version.
What are the main new features?
Four major new features: a 4× more honest model (it flags its own mistakes), an effort control to choose the depth of thinking, a fast mode that is 2.5× faster and 3× cheaper, and dynamic workflows that launch hundreds of sub-agents in parallel in Claude Code.
How much does Claude Opus 4.8 cost?
Pricing is unchanged from Opus 4.7: $5 per million input tokens and $25 per million output tokens. Fast mode is billed at $10 / $50 per million tokens. On subscriptions, the model is included in the claude.ai plans.
Is Opus 4.8 better than GPT-5.5 and Gemini?
On most benchmarks, yes. Opus 4.8 outperforms GPT-5.5 and Gemini 3.1 Pro on agentic coding (69.2% on SWE-Bench Pro), reasoning, computer use and financial analysis. In June 2026, OpenAI no longer even appears in the top 10 models for web development. The only exception: GPT-5.5 keeps the edge on the Terminal-Bench benchmark.
What is the "effort control"?
It is a new setting, next to the model selector in claude.ai and Cowork. It lets you choose how much effort Claude puts into a response. High effort = deeper reasoning for complex tasks. Low effort = faster response and fewer credits used. Available on all plans.
And what is Claude Mythos?
Claude Mythos is a new class of model even more powerful than Opus, which Anthropic is currently reserving for a small number of organisations as part of Project Glasswing (cybersecurity). These models need stronger safeguards before being released. Anthropic plans to make them available to all its customers in the coming weeks.
How do I train my teams on Claude?
AI Sisters supports companies in adopting Claude with tailored, role-based training: an AI Show, hands-on workshops on your real documents, and impact tracking. Our training is Qualiopi-certified and eligible for OPCO funding (French training funds). Your teams master Claude Opus 4.8 and its new tools the very next day.

‍

Where does your company stand with AI? Take the free AI diagnostic in 5 minutes and get your action priorities.