Chat
A conversation with Claude. It is the right place for one-off questions and quick tasks and works best when you do not need standing context.
See also: 3.1 Product Features
Project
A workspace that keeps instructions and reference knowledge together so related chats start with the same context. It fits recurring work for a team or topic.
Exam context: Projects hold what repeats; prompts hold what is one-off.
See also: 3.1 Product Features
Research Mode
A feature for questions that require gathering and synthesizing information from many sources into a cited report. It suits broad investigation, not quick lookups.
Exam context: Pick research mode for multi-source questions that need a synthesized, sourced answer. Verify the citations before relying on them.
See also: 3.1 Product Features
Haiku
The fastest and lightest model tier, well suited to high-volume, simple, or time-sensitive tasks such as classification, short rewrites, and quick formatting.
See also: 3.2 Model Tiers
Sonnet
The balanced model tier, offering strong capability with good speed. It is a sensible default for most everyday business work.
See also: 3.2 Model Tiers
Opus
The most capable model tier, best for hard reasoning, complex analysis, and tasks where quality outweighs speed and cost.
Exam context: Opus is the right choice when the task demands depth. Defaulting to the largest model for everything is usually wrong.
See also: 3.2 Model Tiers
Model Tier
A level in the model lineup that trades speed and cost against depth of capability. Choosing a tier means matching it to the task rather than always choosing the biggest.
See also: 3.3 Cost / Speed / Quality Alignment
Cost / Speed / Quality Trade-off
The balance among how much a run costs, how fast it returns, and how good the result must be. High-volume, low-stakes work favors speed and low cost; high-stakes, complex work favors quality.
Exam context: Match the tier to the task. If a smaller model meets the requirement, the larger one adds cost without benefit.
See also: 3.3 Cost / Speed / Quality Alignment
Context Window
The amount of text Claude can consider at once, including the conversation, uploaded files, and instructions. As a chat grows long, earlier details can be lost or be applied less reliably.
Exam context: Signs of overload include forgotten constraints and contradictions. The fix is to restart with a summary or move standing facts into a Project.
See also: 3.4 Context Limits and Memory
Memory
A way for Claude to carry selected information across conversations. It helps with stable preferences and facts but should not replace explicit Project knowledge for team-critical information.
Product churn: Memory availability and behavior change over time. Rely on the Help Center for current details.
See also: 3.4 Context Limits and Memory
Conversation Summary
A short account of decisions, open questions, and constraints that you carry into a new chat. Read and correct it before use so an error does not carry forward.
See also: 3.4 Context Limits and Memory