What's new in Claude Opus 4.8

Claude Opus 4.8 is Anthropic's most capable generally available model to date. It builds on Claude Opus 4.7. This page summarizes everything new at launch, including fast mode (research preview on the Claude API) and a lower 1,024-token minimum cacheable prompt length.

New model

Claude Opus 4.8 supports the 1M token context window by default on the Claude API, Amazon Bedrock, and Vertex AI (200k on Microsoft Foundry), 128k max output tokens, adaptive thinking, and the same set of tools and platform features as Claude Opus 4.7.

For complete pricing and specs, see the models overview.

New features

Mid-conversation system messages

Claude Opus 4.8 accepts role: "system" messages immediately after a user turn in the messages array (subject to placement rules). This lets you append updated instructions later in a long-running conversation without restating the full system prompt, which preserves prompt cache hits on the earlier turns and reduces input cost on agentic loops. No beta header is required. See Mid-conversation system messages for usage details.

Refusal stop details

The stop_details object on refusal responses (available since Claude Opus 4.7) is now publicly documented. When Claude declines to complete a request, this object describes the category of refusal, in addition to the existing refusal stop reason, making it easier for your application to tell apart different classes of declined request and to route the user to the right next step. No beta header is required. See Handling stop reasons for the category list and handling guidance.

Effort defaults

The effort parameter default on Claude Opus 4.8 is high on all surfaces, including the Claude API and Claude Code. If you set effort explicitly today, your setting is unchanged. See Effort for per-level guidance.

Fast mode

Fast mode is now available for Claude Opus 4.8 as a research preview on the Claude API. Set speed: "fast" to get up to 2.5x higher output tokens per second from the same model at premium pricing. See Fast mode for access, supported models, and pricing.

Lower prompt cache minimum

The minimum cacheable prompt length on Claude Opus 4.8 is 1,024 tokens, lower than on Claude Opus 4.7. Prompts that were too short to cache on Claude Opus 4.7 can now create cache entries with no code changes. See Prompt caching for per-model minimums.

API constraints inherited from Claude Opus 4.7

Sampling parameters not supported

Setting temperature, top_p, or top_k to a non-default value returns a 400 error on Claude Opus 4.8, same as on Claude Opus 4.7. Omit these parameters and use prompting to guide the model's behavior.

Adaptive thinking is the only thinking mode

Like Claude Opus 4.7, Claude Opus 4.8 does not support extended thinking budgets. Setting thinking: {"type": "enabled", "budget_tokens": N} returns a 400 error. Use adaptive thinking and the effort parameter to control thinking depth.

# Before (Opus 4.6 or earlier)
thinking = {"type": "enabled", "budget_tokens": 32000}

# After (Opus 4.7 and later)
thinking = {"type": "adaptive"}
output_config = {"effort": "high"}

Capability improvements

Improvement areas

Compared with Claude Opus 4.7, Claude Opus 4.8 targets behavioral improvements in:

• Long-horizon agentic coding, including better long-context handling, fewer compactions, and better compaction recovery.
• Reasoning effort calibration, with more reliable behavior at each effort level across a range of domains.
• Tool triggering, with fewer cases of skipping a tool call that the task required.

Adaptive thinking

With adaptive thinking enabled, Claude Opus 4.8 triggers reasoning only when it judges the turn needs it. On simple lookups and short agentic steps it responds directly; on complex multi-step problems it reasons before answering. This reduces wasted thinking tokens on bimodal workloads compared to Claude Opus 4.7 at the same effort level. As on Claude Opus 4.7, thinking is off unless you explicitly set thinking: {type: "adaptive"} in your request.

Behavior changes

These are not API breaking changes but may require prompt updates. See Migrating to Claude Opus 4.8 for full guidance.

• Fewer wasted thinking tokens at the same effort level when adaptive thinking is enabled, because the model decides per turn whether to think.
• Better tool triggering. The model is less likely to skip a tool call the task required, an issue some users reported on Claude Opus 4.7.
• Better compaction handling and long-context quality. Long agentic traces stay on task with fewer derailments after compaction.

Migration guide

For step-by-step migration instructions and the full migration checklist, see Migrating to Claude Opus 4.8. If you use Claude Code or the Agent SDK, the Claude API skill can apply these migration steps to your codebase automatically.

Next steps