Claude Opus 4.8

Claude Opus 4.8 is Anthropic’s May 2026 Opus generation that remains active for production compatibility even though Claude Opus 5 is now the current Opus model.

What this model is

Calling a model ‘legacy’ does not mean it has stopped working. Anthropic still lists claude-opus-4-8 as Active (legacy), with retirement not sooner than May 28, 2027. That makes the model relevant for organizations that built prompts, tool schemas and evaluations around 4.8 and need a stable migration window. At the same time, Anthropic explicitly provides migration guidance toward Opus 5, so a new project should treat 4.8 as a compatibility choice rather than the obvious default.

Technical capabilities and model behavior

Claude Opus 4.8 provides a 1M-token context window and up to 128K output tokens. It accepts text and images and produces text. Adaptive thinking is supported with a default high effort, and the model works with Anthropic’s tool and platform ecosystem. Its reliable knowledge cutoff is January 2026. Anthropic also introduced mid-conversation system messages with Opus 4.8, allowing instructions to change during a long session while preserving prompt-cache behavior under the documented placement rules.

How it works in real applications

In production, Opus 4.8 can power long-running coding agents, document analysis and enterprise workflows where model behavior has already been validated. A legal or research system may cache hundreds of thousands of tokens of documents and reuse that context across questions. An engineering agent may depend on particular tool-calling patterns that were tested against 4.8. In those cases, upgrading immediately can create regression risk even if the newer model is generally stronger.

Current status and availability

The model is available through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Anthropic’s lifecycle table keeps it active while identifying it as a legacy generation. This is exactly the kind of model a serious directory should preserve: not the newest, but still operational and important to existing users.

Pricing and deployment considerations

Standard API pricing is $5/M input and $25/M output, with prompt-cache read pricing of $0.50/M and a 50% Batch API discount on input/output. Crucially, those headline prices match Claude Opus 5. That removes the normal cost argument for choosing the older model, so teams should base the decision on validated behavior, migration timing or provider availability rather than price savings.

Who should choose this model?

Choose Opus 4.8 when you already operate it successfully, need time to complete a controlled migration, or have evaluations showing a specific 4.8 behavior that matters to your product. For a greenfield project, start with Opus 5 because Anthropic positions it as the current Opus generation and a step-change improvement on agentic and long-horizon tasks.

Important limitations and trade-offs

The main risk is lifecycle debt. An application can keep working on a legacy model today but still face a future retirement deadline. Its January 2026 internal knowledge also needs tool grounding for current facts. Teams should maintain migration tests, avoid hard-coding undocumented behavior and compare output/tokenization when moving to Opus 5.

Claude Opus 4.8 capabilities and use cases

In addition, its main capabilities include 1M context, 128K output, Adaptive thinking, Vision, Tool use and Structured outputs. For example, common use cases include Legacy enterprise agents, Long documents, Coding, Migration validation and Complex analysis.

Who should consider Claude Opus 4.8?

In practice, this model may suit Existing Opus 4.8 integrations, Long-context enterprise workflows, Agentic coding, Migration testing and Stable legacy behavior. Also, notable strengths include Still active, 1M context, Large output limit and Broad platform support. However, review trade-offs such as January 2026 knowledge cutoff, Anthropic recommends migration to Opus 5 and Maintaining an older model can increase long-term migration work before adopting it.

Claude Opus 4.8 pricing and access

Meanwhile, $5 per 1M input tokens and $25 per 1M output tokens. Prompt cache reads cost $0.50/M and Batch API input/output receives a 50% discount. Opus 4.8 costs the same $5/M input and $25/M output as Opus 5, so new projects should generally prefer Opus 5 unless compatibility or evaluation results justify 4.8.

Official resources and verification

Use the official model website, official documentation, pricing or release source and additional primary source to confirm current availability, limits and pricing. Product details can change after publication, so rely on primary documentation for final decisions.

Compare with other AI models

Next, continue your research in the AI models directory, Anthropic models and General Purpose Language Model models. Compare providers, pricing, modalities and practical limitations side by side to choose the right model for your workflow.