What happened
In May 2026, on May 28, Anthropic launched Claude Opus 4.8, a new version of what was then its most advanced general-access model. It builds on Opus 4.7 with benchmark gains and arrived at the same price: $5 per million input tokens and $25 per million output tokens.
It was available the same day everywhere: the Claude apps, the Claude API (claude-opus-4-8), Amazon Bedrock, Google Cloud and Microsoft Foundry.
What improved
Anthropic itself called Opus 4.8 a "modest but tangible" improvement. The highlights:
- More honesty: according to Anthropic, the model is about 4 times less likely than its predecessor to let flaws in code it wrote pass unremarked. Testers reported it flags uncertainty more and makes fewer unsupported claims.
- Better judgment on agentic tasks: it asks more questions, catches its own mistakes and pushes back on weak plans before big changes.
- Computer and browser use: one partner reported 84% on Online-Mind2Web, above Opus 4.7 and GPT-5.5.
- Alignment: internal assessment showed rates of misaligned behavior (such as deception or cooperating with misuse) substantially lower than Opus 4.7.
Launched alongside
- Dynamic workflows (research preview) in Claude Code: Claude plans the work and runs hundreds of parallel subagents in a single session, verifying results before reporting back. Anthropic cites codebase-scale migrations across hundreds of thousands of lines, from kickoff to merge. Available on Enterprise, Team and Max plans.
- Effort control in claude.ai and Cowork, on all plans: more effort means better answers; less effort means faster replies and slower use of rate limits.
- Fast mode (up to 2.5x faster) became three times cheaper than on previous models: $10 / $50 per million tokens.
- The Messages API now accepts system entries mid-conversation, so developers can update an agent's instructions, permissions or token budget without breaking the cache.
Opus 4.8 defaults to high effort. For hard tasks and long workflows, Anthropic recommends "extra" (xhigh in Claude Code).
What came next
In the announcement, Anthropic already said it planned to release "in the coming weeks" a new class of models above Opus, the Mythos class, once its cyber safeguards were strong enough. That happened on June 9 with Claude Fable 5, covered in another post in this series.
What this means for businesses
- Low-risk upgrade: same price, same API. Test Opus 4.8 on your current workflows and compare.
- Honesty matters in automation: a model that admits uncertainty reduces silent errors in code and reports.
- Parallel agents need controls: with hundreds of subagents, automated tests and merge review are essential.
- Match effort to cost: use high effort only where quality pays off.
Sources
- Anthropic, "Introducing Claude Opus 4.8" (2026-05-28): https://www.anthropic.com/news/claude-opus-4-8
- Model documentation: https://platform.claude.com/docs/en/models/opus-4-8/overview
- System card: https://www.anthropic.com/claude-opus-4-8-system-card
Want to find out which AI model best fits your company's workflows? KGA IT offers a free consultation: kga-it.com.