Claude 3 Models

The Claude 3 family launched on Mar 4, 2024 with Claude 3 Opus, Sonnet and Haiku, followed by Claude 3.5 Sonnet (Jun 21, 2024), its upgrade and Claude 3.5 Haiku (Oct 22, 2024), and Claude 3.7 Sonnet (Feb 24, 2025), the first Claude with extended thinking. All of them are retired on the Claude API; the last, Claude 3 Haiku, on Apr 20, 2026.

The Claude 3 family was announced on Mar 4, 2024 as three models, Claude 3 Opus, Claude 3 Sonnet and Claude 3 Haiku, the first time Anthropic used those names for large, middle and small (Introducing the next generation of Claude, Anthropic, Mar 4, 2024). Claude 3.5 Sonnet, 3.5 Haiku and 3.7 Sonnet followed over the next year, and 3.7 Sonnet introduced extended thinking. Every Claude 3 model is now retired on the Claude API: the last, Claude 3 Haiku, stopped answering on Apr 20, 2026.

Quick facts

First release	Claude 3 Opus and Sonnet, Mar 4, 2024
Last release	Claude 3.7 Sonnet, Feb 24, 2025
Context window	200K tokens on every Claude 3 model
Launch prices	Opus $15 / $75, Sonnet $3 / $15, Haiku $0.25 / $1.25 per million tokens
First with image input	Claude 3 (all three models)
First with extended thinking	Claude 3.7 Sonnet
Status	All retired on the Claude API (as of Nov 6, 2026)
Never released	Claude 3.5 Opus

Sources: Introducing the next generation of Claude, Anthropic, Mar 4, 2024; Claude 3.7 Sonnet and Claude Code, Anthropic, Feb 24, 2025; Model deprecations, Anthropic docs, read Nov 6, 2026.

Every Claude 3 model

Model	Released	API id	Launch price (input / output)	Retired
Claude 3 Opus	Mar 4, 2024	`claude-3-opus-20240229`	$15 / $75	Jan 5, 2026
Claude 3 Sonnet	Mar 4, 2024	`claude-3-sonnet-20240229`	$3 / $15	Jul 21, 2025
Claude 3 Haiku	Mar 13, 2024	`claude-3-haiku-20240307`	$0.25 / $1.25	Apr 20, 2026
Claude 3.5 Sonnet	Jun 21, 2024	`claude-3-5-sonnet-20240620`	$3 / $15	Oct 28, 2025
Claude 3.5 Sonnet (upgraded)	Oct 22, 2024	`claude-3-5-sonnet-20241022`	$3 / $15	Oct 28, 2025
Claude 3.5 Haiku	Oct 22, 2024	`claude-3-5-haiku-20241022`	$0.80 / $4 (from Dec 3, 2024)	Feb 19, 2026
Claude 3.7 Sonnet	Feb 24, 2025	`claude-3-7-sonnet-20250219`	$3 / $15	Feb 19, 2026

Prices per million tokens. Sources: launch posts linked below; Model deprecations, Anthropic docs, read Nov 6, 2026.

Claude 3 Opus, Sonnet and Haiku (Mar 2024)

All three took images (photos, charts, diagrams) as well as text and had a 200K-token context window, with more than 1 million tokens offered to selected customers (Introducing the next generation of Claude). Anthropic reported over 99% recall for Opus on its Needle In A Haystack test, and said Sonnet was about twice as fast as Claude 2 and 2.1. Sonnet powered free claude.ai and Opus came with Claude Pro; the API became generally available in 159 countries.

A table with columns Claude 3 Opus, Claude 3 Sonnet and Claude 3 Haiku (outlined in orange), GPT-4, GPT-3.5, Gemini 1.0 Ultra and Gemini 1.0 Pro. Opus scores 86.8% on MMLU, 50.4% on GPQA Diamond, 95.0% on GSM8K, 60.1% on MATH, 90.7% on MGSM, 84.9% on HumanEval, 83.1 on DROP, 86.8% on BIG-Bench-Hard, 96.4% on ARC-Challenge and 95.4% on HellaSwag, its scores in green. — Anthropic's Claude 3 launch table compares Opus, Sonnet and Haiku with GPT-4, GPT-3.5 and Gemini 1.0 on ten benchmarks. Anthropic

0:13 Haiku, released on Mar 13, 2024, read about 21K tokens a second on short prompts, and Anthropic said one dollar bought analysis of 400 Supreme Court cases or 2,500 images. All three were released at AI Safety Level 2 under Anthropic's Responsible Scaling Policy.

Anthropic posted "An update on our model deprecation commitments for Claude Opus 3" on Feb 25, 2026, after its Jan 5 retirement (Models, Anthropic, read Nov 6, 2026); that post was not read for this page.

Claude 3.5 Sonnet (Jun 21, 2024)

0:49 Claude 3.5 Sonnet ran twice as fast as Claude 3 Opus at the Sonnet price of $3 and $15, and solved 64% of Anthropic's internal agentic coding problems against 38% for Claude 3 Opus. It launched with Artifacts in preview on Claude.ai and was free on the website and iOS app. The UK AI Safety Institute tested it before release. The same post promised Claude 3.5 Opus later in 2024; no such model appears in Anthropic's list of releases.

A chart with Intelligence (benchmark scores) up the side and Cost (price per million tokens) along the bottom. A dashed curve rises from Claude 3 Haiku through a faded Claude 3 Sonnet to Claude 3 Opus; an arrow rises straight up from Claude 3 Sonnet to Claude 3.5 Sonnet, which sits higher than Opus at Sonnet's cost. — Anthropic's chart of intelligence against cost shows Claude 3.5 Sonnet above Claude 3 Opus at the price of Claude 3 Sonnet. Anthropic

The upgraded 3.5 Sonnet, 3.5 Haiku and computer use (Oct 22, 2024)

0:54 An upgraded Claude 3.5 Sonnet raised its SWE-bench Verified score from 33.4% to 49.0% and introduced Computer Use in public beta: the model could look at screenshots and move a cursor, which Anthropic itself called "at times cumbersome and error-prone". It scored 14.9% on OSWorld from screenshots alone, against 7.8% for the next best model then. Claude 3.5 Haiku, text-only at launch, matched Claude 3 Opus on many tests; the post was updated on Dec 3, 2024 to give its price as $0.80 and $4.

A table with columns Claude 3.5 Sonnet (new) and Claude 3.5 Haiku outlined in orange, then Claude 3.5 Sonnet, GPT-4o, GPT-4o mini, Gemini 1.5 Pro and Gemini 1.5 Flash. The new 3.5 Sonnet scores 65.0% on GPQA Diamond, 78.0% on MMLU Pro, 93.7% on HumanEval, 78.3% on MATH, 16.0% on AIME 2024, 70.4% on MMMU, 49.0% on SWE-bench Verified (against 33.4% for the earlier 3.5 Sonnet) and 69.2% retail and 46.0% airline on TAU-bench. A footnote says the tables exclude OpenAI's o1 models. — Anthropic's October 2024 table compares the upgraded Claude 3.5 Sonnet and Claude 3.5 Haiku with the first 3.5 Sonnet, GPT-4o and Gemini 1.5 models. Anthropic

Claude 3.7 Sonnet and hybrid reasoning (Feb 24, 2025)

0:04 Claude 3.7 Sonnet was, in Anthropic's words, "the first hybrid reasoning model on the market": one model that answered at once or, with extended thinking on, reasoned step by step first, with a thinking budget up to its 128K-token output limit. Thinking tokens were billed as output. It scored 70.3% on SWE-bench Verified with Anthropic's high-compute setup and 63.7% without. It reached every plan, Free included, though extended thinking needed a paid plan. Claude Code launched alongside it as a research preview. See Extended and Adaptive Thinking for how thinking works today.

Frequently asked questions

Can I still use Claude 3 Opus? No. Claude 3 Opus was retired on the Claude API on Jan 5, 2026, and requests to it fail. Anthropic says it keeps the weights of retired models and hopes to make them available again some day.

When was Claude 3.5 Sonnet retired? Both versions of Claude 3.5 Sonnet were retired on the Claude API on Oct 28, 2025, after being deprecated on Aug 13, 2025.

What replaced Claude 3.7 Sonnet? Claude 3.7 Sonnet was retired on Feb 19, 2026. Its line continued through Claude Sonnet 4, 4.5, 4.6 and 5 to Claude Sonnet 5.5, Anthropic's current Sonnet as of Nov 6, 2026.

Did Anthropic release Claude 3.5 Opus? No. Anthropic said in Jun 2024 that Claude 3.5 Opus would come later that year, but it never shipped; the next Opus was Claude Opus 4 in May 2025.

What was the context window of Claude 3? Every Claude 3, 3.5 and 3.7 model had a 200K-token context window, about 150,000 words. Current Claude models accept 1M tokens.