Claude¶

Development line: project:claude · thread claude-product-development
Last event: 2023-11-22 · 2 dated since 2023-05-12 · Researched: 2026-09-04 · confidence: high

What it is¶

Claude is Anthropic’s model family for chat users, application developers, and software teams.

• Text and images: takes text and images, and returns text.
• Capabilities: supports multilingual tasks, vision, tool use, and long-context workflows.
• Surfaces: runs through chat, the Messages API, and Claude Code for repository work.

Development line¶

• 2023-05-12 — Claude expanded to 100K-token context windows. Claude grew from 9K to 100K tokens in the 100K-context release so prompts could hold whole documents and full codebases. The official article is dated 2023-05-11.
• 2023-11-22 — Claude 2.1 introduced a 200K-token context window. Claude 2.1 raised the limit to 200K tokens, added system prompts and beta tool use, and cut false statements. The official article is dated 2023-11-21.

What changed¶

Claude — the dated development line:

• 2023-05-12: The 100K-context release expanded Claude from 9K to 100K tokens so prompts could hold whole documents and codebases. The official article is dated 2023-05-11.
• 2023-11-22: Claude 2.1 raised the limit to 200K tokens, added system prompts and beta tool use, and lowered false-statement rates. The official article is dated 2023-11-21.
• 2026-09-04 (found today): Current documentation describes a model family with text-and-image input, tool use, model-specific limits, and a 200K–1M-token context range.

How to use this¶

From 2023-05-12, practitioners could consider Claude for workflows requiring up to 100K tokens of context. From 2023-11-22, Claude 2.1 extended that planning envelope to 200K tokens.

1. Choose the surface: use Claude.ai for chat, the Console and API for an application, or Claude Code for a repository task. — https://docs.anthropic.com/en/docs/about-claude/models/overview
2. For API integration, create a Claude Console account and API key, export it as ANTHROPIC_API_KEY, then install the supported SDK. — https://docs.anthropic.com/en/docs/get-started
3. Choose a current model against capability, latency, cost, effort, and your task evaluation before committing to production. — https://platform.claude.com/docs/en/about-claude/models/choosing-a-model
4. Make a Messages API call with the selected model, a max_tokens limit, and the user message; add tools, files, or context-management features only when the task needs them. — https://docs.anthropic.com/en/docs/get-started
5. For codebase work, install Claude Code on a supported surface, start it in the project directory, and review the file and command actions it performs. — https://code.claude.com/docs/en/overview

Best practices¶

• Define specific, measurable success criteria and an evaluation set before changing prompts or models. — https://platform.claude.com/docs/en/test-and-evaluate/develop-tests
• State the task, context, constraints, output format, and ordered steps directly. Add relevant examples when output consistency matters. — https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
• For inputs above 20K tokens, place long documents before the query, structure multiple documents clearly, and ask for supporting quotations before synthesis. — https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
• Treat model choice and effort as performance controls alongside prompting; use evaluations to trade capability against latency and cost. — https://platform.claude.com/docs/en/about-claude/models/choosing-a-model
• Pin model IDs for reproducibility and monitor their lifecycle. Dateless IDs from the 4.6 generation onward are fixed snapshots, not moving aliases. — https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions

Superseded by this¶

• 2023-05-12: Treating 100K tokens as Claude’s operational context ceiling is obsolete; current models range from 200K to 1M input tokens.
• 2023-11-22: Deploying Claude 2.1 guidance is obsolete; API model claude-2.1 was retired on 2025-07-21.
• 2023-11-22: Treating tool use as a beta-only capability is obsolete; current-model documentation lists tool use as a supported capability.
• Pre-4.6 alias assumptions are obsolete for current dateless model IDs: each current ID identifies a fixed model snapshot.

Still unknown¶

• The two dated events are not a complete release chronology between 2023 and today; the current documentation is used only as a live snapshot.
• The supplied event dates are one day later than the official article dates: 2023-05-12 versus 2023-05-11, and 2023-11-22 versus 2023-11-21.
• Plan availability, region, platform, and pricing can differ by account and change after this snapshot; verify them in the Console before rollout.

Sources¶

Agent brief¶

• Subject: project:claude, thread claude-product-development, 2 dated events 2023-05-12 → 2023-11-22.
• Practical note: From 2023-05-12, practitioners could consider Claude for workflows requiring up to 100K tokens of context; from 2023-11-22, Claude 2.1 extended that planning envelope to 200K tokens.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.