Claude 3
Era 5 · Generative AI, reasoning, and agents (2022–2026) › Frontier models
- Lineage
- Neurons and networks
- Kind
- Implementation
Who
Anthropic
What happened
Anthropic presents Claude 3, a family of three models of increasing capability: Haiku, Sonnet, and Opus. All three accept images as well as text and have a 200,000-token context window.
Why it mattered
A family in three sizes that lets users choose between speed, cost, and capability.
The full story
On March 4, 2024, Anthropic presented Claude 3 as a family of three models. Haiku was the fastest and cheapest, designed to respond almost instantly; Sonnet balanced intelligence and speed; and Opus was the most capable.
The split reflects the fact that not every task needs the same thing: some call for speed and low cost, others for maximum capability.
All three accepted images, such as photos, charts, and technical diagrams, and had a context window of 200,000 tokens. Anthropic stated that Opus outperformed its rivals on most common benchmarks and that it retrieved a fact hidden in a very long text with more than 99 percent accuracy, the test known as the "needle in a haystack."
Opus and Sonnet were available that day in 159 countries through the API. Sonnet powered the free version of claude.ai, and Opus was for subscribers. Haiku was announced for later.
The three-size scheme held in the generations that followed, although the largest model was not always the best: in June 2024 Claude 3.5 Sonnet outperformed Claude 3 Opus. Artifacts arrived with it.
Sources
Scholarship and context
Further reading
How to cite
“Claude 3.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/claude-3
This entry was last reviewed on November 13, 2026.