What are the current Claude models?

As of November 2026, Anthropic's model overview lists Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. All four have a 1 million token context window and up to 128,000 output tokens.

When did Claude get a 1 million token context window?

Claude 2 started at 100,000 tokens, and Claude 2.1 and Claude 3 used 200,000. A 1 million token window became standard with Claude Opus 4.6 in February 2026 and is standard across the Claude 5 family.

What is the difference between extended thinking and adaptive thinking?

Extended thinking, introduced with Claude 3.7 Sonnet, let developers set a thinking budget by hand. Adaptive thinking, from Opus 4.6 onwards, lets the model decide how much to think, guided by an effort setting.

When did Claude Code come out?

Claude Code arrived as a limited research preview alongside Claude 3.7 Sonnet in February 2025 and became generally available with Claude 4 in May 2025.

Developer blog

Claude 1 to Claude 5: What Actually Changed, in Plain English

Published 4 November 2026 · 4 min read

By Cong Le · Pocket Grove

Original research brief: 4 November 2026. Sources reviewed: 4 November 2026.

Quick answer: Claude went from a chat box (2023), to reading very long documents (Claude 2), to seeing images (Claude 3), to building things beside the chat and using a computer (Claude 3.5), to thinking before it answers (3.7), to agents that code for long stretches (Claude 4 and Claude Code), to today's Claude 5 family, where you mostly choose an effort level. The context window grew from 100K to 1M tokens along the way.

This is the Anthropic version of our GPT-2 to GPT-6 explainer. Every step links to Anthropic's own announcement or docs.

2023: Claude 1 and Claude 2, long documents

Claude launched in March 2023, in two flavours: Claude, and a faster, cheaper Claude Instant.

Claude 2 (July 2023) was the first to stand out. It could take up to 100,000 tokens, roughly a short book, in one go, and it opened the public claude.ai site. Claude 2.1 (November 2023) doubled that to 200,000 tokens and added system prompts, plus an early beta for calling your own tools.

How you used it: paste in a long document and ask questions about it.

2024: Claude 3 and 3.5, seeing, building and clicking

Claude 3 (March 2024) introduced the three sizes people still know: Haiku (fast and cheap), Sonnet (balanced) and Opus (smartest). It could also read photos, charts and diagrams.

Claude 3.5 Sonnet (June 2024) brought Artifacts: code, documents and designs open in a panel beside the chat. In October 2024 Anthropic released a public beta of computer use: Claude looks at a screen, moves the mouse, clicks and types. Anthropic itself called it experimental and error-prone.

How you used it: pick a size, show it pictures, watch it build something.

2025: Claude 3.7 and Claude 4, thinking and agents

Claude 3.7 Sonnet (February 2025) added extended thinking: the model reasons before it answers, and developers set a thinking budget. The same launch previewed Claude Code, a coding agent that runs in your terminal.

Claude 4 (May 2025) made Claude Code generally available, with VS Code and JetBrains support. Claude could use tools such as web search while thinking, and keep memory files between tasks when given file access.

Later in the year, Opus 4.5 (November 2025) added an effort control and automatic summarising, so long conversations stop running out of room.

How you used it: hand it a task, let it think and work, review the result. Our Claude Code best practices post covers this style of work.

2026: 1M context and the Claude 5 family

Opus 4.6 (February 2026) made a 1 million token context window standard and introduced adaptive thinking: instead of you setting a thinking budget, the model decides how much to think, guided by effort.

The Claude 5 family followed from June 2026. As of November 2026, Anthropic's models overview lists four current models, all with 1M context and up to 128K output:

Model	Price per million tokens (input / output)	Default effort
Fable 5.1	$10 / $50	high
Opus 5.5	$4 / $20	medium
Sonnet 5.5	$2 / $10	high
Haiku 5.5	from $0.10 / $0.50	medium

Fable 5.1 (September 2026) can change effort mid-conversation (beta) and gives progress updates between tool calls. Opus 5.5 performs at Fable 5.1's level on most work, costs 40% less to run than Opus 5, and always thinks: you can no longer turn thinking off.

How you use it: choose a model and an effort level, and let it decide how hard to think.

What this means for you

The way you use it changed more than how clever it is. Chat, then vision, then side panels and computer use, then thinking, then agents. Each step changed how you work with it.
Context is rarely the bottleneck now. 1M tokens fits a whole codebase. Filling the window is still slower and costs more, so curate it. See context engineering explained.
Old models retire. Claude 1, 2, 3 and 3.5 are all retired. Pin model IDs in your apps and plan upgrades.

Compare with OpenAI's GPT-2 to GPT-6 and Google's BERT to Gemini 4.

Frequently asked questions

What are the current Claude models?: As of November 2026, Anthropic's model overview lists Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. All four have a 1 million token context window and up to 128,000 output tokens.
When did Claude get a 1 million token context window?: Claude 2 started at 100,000 tokens, and Claude 2.1 and Claude 3 used 200,000. A 1 million token window became standard with Claude Opus 4.6 in February 2026 and is standard across the Claude 5 family.
What is the difference between extended thinking and adaptive thinking?: Extended thinking, introduced with Claude 3.7 Sonnet, let developers set a thinking budget by hand. Adaptive thinking, from Opus 4.6 onwards, lets the model decide how much to think, guided by an effort setting.
When did Claude Code come out?: Claude Code arrived as a limited research preview alongside Claude 3.7 Sonnet in February 2025 and became generally available with Claude 4 in May 2025.