New: Claude Opus 4.8 Is Now EU-Hosted in Corporate LLM

On May 28, 2026, Anthropic released Claude Opus 4.8, its strongest generally available model to date. It is available in Corporate LLM starting today, EU-hosted in the model picker, in the same plan, at no extra cost. The concrete difference from Opus 4.7 lies where the model thinks across many steps: multi-step, agentic tasks and more honest, more reliable answers.

Overview: Claude Opus 4.8 is now EU-hosted and GDPR-compliant in the Corporate LLM model picker. Anthropic released it on May 28, 2026, six weeks after Opus 4.7. It is more reliable on multi-step, agentic tasks and gives confident-sounding wrong answers less often, runs under the same data processing agreement (DPA) as the other standard models, and costs nothing extra. Opus 4.7 remains available in parallel.

Where Claude Opus 4.7 hit its limits in everyday business

Opus 4.7 was the strongest model in the picker for weeks, and for most SMB tasks it still suffices. Three points nevertheless created friction in practice:

• Agentic multi-step tasks break off. When a model plans a task through several steps on its own — researching, checking, correcting — every intermediate error counts double. On long chains, that is exactly the sore spot.
• Confident-sounding wrong answers. If you used 4.7 for demanding analyses, you occasionally got a convincingly phrased but wrong answer. In a business context, that is more expensive than an honest "this is unclear."
• Verification overhead on everything that matters. The consequence: for a contract analysis or a quote calculation, you preferred to double-check every statement yourself. Which eats up the time savings.

What is the difference between Claude Opus 4.8 and 4.7?

According to Anthropic, Opus 4.8 targets exactly these three points. The substantiated numbers from the official Anthropic release:

• Sees multi-step tasks through. On Online-Mind2Web, a benchmark for agentic tasks across many steps, 4.8 scores 84 percent — the highest value Anthropic has measured. For you, that means: in chat, the model breaks off mid-way through long, nested requests less often.
• More honest instead of confidently wrong. Anthropic measured this in code review: 4.8 lets an error pass without comment roughly four times less often than 4.7. Anthropic calls this a step toward "honesty as a feature," and the same caution helps with every statement that has to be correct.
• Agentic tasks go the distance. On the Legal Agent Benchmark, 4.8 achieves the highest score ever measured and is the first model to crack the 10 percent mark under the strict all-pass standard.

As a supplementary tier-2 figure from the trade press: on SWE-Bench Pro, 4.8 rises to 69.2 percent, up from 64.3 percent for 4.7. This number does not come directly from the Anthropic release, but it fits the picture: stronger at real-world code work.

Claude Opus 4.8 for SMBs: everyday use cases

Benchmarks are one thing. Here is where the improvements actually land in everyday SMB work, right in the normal Corporate LLM chat:

• Quotes and technical documents you can rely on more. Whether it is a customer-specific machinery quote, technical documentation, or a client report: 4.8 scrutinizes its own statements more critically and gives a confident-sounding but wrong answer less often. In practice, that means fewer incorrect specs, numbers, or prices that only surface at the customer.
• Multi-step expert tasks in a single conversation. Review a contract, list the risks, draft a counterproposal: all in one pass. Because 4.8 sustains long chains of reasoning more reliably, exactly those multi-step requests that 4.7 sometimes failed at break off mid-way less often.
• Complex research and strategy questions. When a question requires multiple sources, trade-offs, and intermediate steps, 4.8 delivers the more considered result instead of losing the thread halfway. For the demanding 20 percent of tasks, you select 4.8 in the picker; for routine work, the cheaper models stay active.

Is Claude Opus 4.8 GDPR-compliant and EU-hosted?

Opus 4.8 can be used in a GDPR-compliant way in Corporate LLM: it runs on the EU-hosted Anthropic endpoint, documented via the sub-provider list in the data processing agreement (DPA). Three guarantees apply, as with every standard model: EU hosting as the default, no training on your data, no persistence at the provider after the response is generated. The EU status is visible in the chat header, so every conversation remains transparent. For data protection at mid-sized companies, that means: the same level of protection as with the other standard models, with no extra effort.

That is the real reason 4.8 was available on release day: as soon as a new model runs on EU infrastructure, our engineering team adds it to the picker. You do not have to do anything.

How much does Claude Opus 4.8 cost in Corporate LLM?

Claude Opus 4.8 is in the model picker starting today, in every paid plan (Starter, Pro, Max), with no surcharge for the model itself. Only the free tier does not have it. You pay your usual per-seat price per month; Opus 4.8 draws on your plan budget like the other top models, and you can see the utilization in the dashboard. Opus 4.7 remains active in parallel for now, in case you deliberately want to stay on the previous version. You can switch existing conversations to 4.8 via the model switcher in the header without losing context.

When Claude Opus 4.8 is not the best choice

Even the best model is not the right one for every task. Three cases where you deliberately choose differently:

• Pure routine and cost optimization. For email drafts and short summaries, Gemini 3 Flash-Lite or Mistral Small are significantly cheaper. Using Opus 4.8 for standard requests is wasted budget.
• Searching very long documents. For pure long-context retrieval across the full 1M-token span, Gemini 3 Pro in the picker is often ahead. That applies to the entire Opus line, not just 4.8.
• Strictly EU-only without exception. Opus 4.8 is EU-hosted, so it is unproblematic. But if you need a non-EU premium model for a special case, you enable it deliberately per team instead of making it the default.

Try Claude Opus 4.8 EU-hosted now

Open Corporate LLM, click the model picker at the top, and select Claude Opus 4.8. The EU badge confirms the hosting location directly in the header.

More background: why Corporate LLM bundles GPT, Claude, Gemini, and Mistral in one platform.

No access yet? Create a free account.