New Model Focuses on Smarter Task Execution and Developer Control
Anthropic today launched Claude Opus 4.8, its latest AI model, bringing substantial agentic improvements, new dynamic workflows, and a significantly faster "fast mode" at unchanged pricing. The update focuses on enhancing the AI's ability to handle complex tasks with greater autonomy and efficiency.
Claude Opus 4.8 delivers notable advancements in agentic capabilities, allowing the model to perform longer autonomous runs and deeper reasoning. This translates into a more robust AI that can navigate obstacles, recover from its own errors, and discern when to seek assistance versus proceeding independently. For coding, Opus 4.8 reads codebases like an engineer, plans edits, and maintains context across extended sessions in real repositories.
The new model also introduces a "fast mode" for Claude Opus 4.8, which operates at 2.5 times the speed of previous models. Crucially, this accelerated mode is now three times cheaper, making high-speed processing more accessible. Opus 4.8 also employs adaptive thinking, triggering reasoning only when necessary, which reduces wasted thinking tokens compared to Opus 4.7 at the same effort level.
Anthropic has rolled out several new features designed to give users and developers more control and flexibility. A key addition is "dynamic workflows" in Claude Code, available as a research preview. This feature empowers Claude to tackle very large-scale problems by running hundreds of parallel subagents in a single session and verifying its work before reporting back.
high across all surfaces, including Claude Code and the Messages API, ensuring thorough task completion. For developers, the Messages API now accepts system entries inside the messages array, enabling mid-conversation instruction updates without disrupting the prompt cache.stop_details when Claude Opus 4.8 declines a request. This allows applications to route different classes of refusals to appropriate next steps. Claude Opus 4.8 maintains a 1M token context window by default and 128k max output tokens, alongside support for high-resolution image input up to 2576 pixels on the long edge. The minimum cacheable prompt length for prompt caching has also been lowered to 1,024 tokens, an improvement over Opus 4.7.Claude Opus 4.8 is available globally starting today across multiple platforms. Users can access the new model on claude.ai, the Claude Platform, and all major cloud platforms, including Amazon Bedrock and Claude Platform on AWS. This broad availability ensures that a wide range of users, from individual creators to enterprise developers, can leverage the latest advancements.
Anthropic has maintained the pricing structure for Opus 4.8, keeping it consistent with previous versions. The model is priced at $5 per million input tokens and $25 per million output tokens. This decision underscores Anthropic's commitment to delivering enhanced capabilities without increasing the cost for its users.