Claude's latest long-running features enhance AI usability for developers, enabling more complex interactions and improved context retention.
Long-running Claude tasks enable autonomous AI operations lasting hours, transforming complex multi-step workflows from human-shepherded iterations into single-submission jobs.
Signal analysis
Anthropic has introduced long-running capabilities for Claude, enabling AI operations that extend from minutes to hours rather than the traditional seconds-to-minute response windows. This release transforms Claude from a request-response assistant into a system capable of sustained, autonomous work on complex tasks. The feature is available through a new async API and Claude's enterprise interface.
The technical implementation separates task submission from result retrieval. Developers submit a task description and receive a task ID immediately. The task executes in Anthropic's infrastructure with optional progress callbacks, and results are retrieved when complete or at any point to check status. This architecture supports tasks requiring substantial iteration, research, or multi-step processing.
Initial supported durations range from 5 minutes to 4 hours, with longer durations planned for future releases. Within these windows, Claude can execute multi-step plans, perform web research through connected tools, generate comprehensive documents, and iterate on complex problems without timeout constraints. Pricing follows a time-based model in addition to standard token pricing.
Research teams conducting comprehensive analyses benefit most from long-running Claude. Tasks previously requiring multiple prompt iterations - like competitive analysis across dozens of sources or systematic literature review - can be submitted as single requests. Claude autonomously gathers information, synthesizes findings, and produces comprehensive outputs without human shepherding through each step.
Software teams needing large-scale code generation or refactoring can leverage extended execution. Generating complete test suites for existing codebases, performing comprehensive code reviews, or migrating deprecated dependencies across large projects become feasible as single tasks. The extended time window allows Claude to iterate until quality thresholds are met rather than producing rushed initial outputs.
Organizations with limited AI engineering resources gain leverage. Previously, sustained AI workflows required orchestration systems that chained multiple Claude calls. Long-running tasks encapsulate this orchestration within Anthropic's infrastructure, reducing the need for custom pipeline development. Smaller teams can achieve sophisticated automation without dedicated AI engineering.
Getting started requires the Claude Enterprise or Teams subscription with long-running tasks enabled. Access requires API key permissions - check your API key settings in the Anthropic Console under Features. Install the latest SDK: `pip install anthropic>=0.40.0` or `npm install @anthropic-ai/[email protected]`. The async methods are in a separate submodule.
Submit a task using `client.tasks.create(prompt=your_prompt, max_duration_minutes=60)`. The response includes a task_id immediately. To check status: `client.tasks.retrieve(task_id)` returns status (queued, running, completed, failed) and partial results if available. For completion: poll periodically or use webhooks by providing a callback_url during creation.
Crafting effective long-running prompts differs from standard prompts. Include explicit success criteria, anticipated obstacles, and instructions for iteration. Example: 'Research X. Success criteria: minimum 10 sources, findings synthesized, contradictions resolved. If initial search insufficient, expand search terms. Continue until criteria met or 2 hours elapsed.' The extended execution enables Claude to act on these meta-instructions.
OpenAI's Assistants API provides similar persistent context capabilities but operates differently. Assistants maintain conversation threads across multiple synchronous calls, while Claude's long-running tasks execute autonomously without intermediate interaction. For workflows requiring human checkpoints, Assistants' thread model may be preferable. For fully autonomous operations, Claude's approach reduces integration complexity.
Cost models differ substantially. OpenAI Assistants charges per token across all thread messages. Claude's long-running tasks add time-based charges but may process fewer total tokens for self-correcting tasks where iteration happens internally rather than through multiple visible exchanges. Estimate costs for specific workflows rather than assuming one model is universally cheaper.
Tool integrations are more comprehensive in Claude's long-running context. The extended execution window allows sequential tool calls without timeout pressure, enabling workflows that gather information from multiple sources, synthesize, then gather more. Assistants' tools execute within standard API timeouts, limiting complex multi-tool choreography.
Anthropic's roadmap extends maximum task duration to 12 hours by Q4 2026, enabling full-workday autonomous operation. This opens possibilities for overnight processing jobs - submit a research task at end of day, review comprehensive results the next morning. The infrastructure supporting this requires durable execution guarantees that Anthropic is actively developing.
Integration with enterprise systems is expanding. Upcoming releases will allow long-running tasks to authenticate with corporate systems using OAuth tokens, enabling Claude to directly access internal APIs and databases during extended execution. This positions Claude as an autonomous agent within enterprise workflows rather than a disconnected AI service.
The broader trajectory points toward AI workers rather than AI assistants. Long-running tasks represent the transition from AI that responds to requests to AI that performs jobs. Organizations should prepare for workflow restructuring as AI capabilities expand from assistance to execution.
Watch the breakdown
Prefer video? Watch the quick breakdown before diving into the use cases below.
Best use cases
Open the scenarios below to see where this shift creates the clearest practical advantage.
One concise email with the releases, workflow changes, and AI dev moves worth paying attention to.
More updates in the same lane.
Cursor introduces self-hosted cloud agents, empowering developers with flexibility and control over their AI tools. Discover how this innovation can transform your development workflow.
Cursor's Warp Decode feature enhances AI-driven code interpretation, streamlining development workflows and improving productivity for developers. Discover how this innovation reshapes coding practices.
Together AI has announced the general availability of Instant Clusters, a new feature that streamlines AI model training and deployment. This innovative tool promises to enhance productivity and collaboration among developers working on AI projects.
