Model Tracker
Claude 3 Haiku
Claude 3 Haiku is an AI model made by Anthropic. Anthropic released it on 13 March 2024. Anthropic retired it on 20 April 2026.
In inglese
Claude 3 Haiku is an AI model made by Anthropic. Anthropic released it on 13 March 2024. Anthropic retired it on 20 April 2026. It takes text, image and returns text. Its context window is 200,000 tokens. Through the API it costs $0.25 per 1M tokens for input and $1.25 per 1M tokens for output. Its knowledge cutoff is August 2023.
- Released
- 13 March 2024 — www.anthropic.com, read 16 October 2026
- Status
- Not confirmed: two readings disagree: retired and ga
- Takes
- Text, Image — assets.anthropic.com, read 16 October 2026
- Returns
- Text — assets.anthropic.com, read 16 October 2026
- Open weights
- No — alignment.anthropic.com, read 16 October 2026
- Licence
- Not confirmed: The maker's pages reviewed do not name a licence for Claude 3 Haiku.
- Context window
- 200,000 tokens — www.anthropic.com, read 16 October 2026
- Longest answer
- Not confirmed: The maker's pages reviewed do not print Claude 3 Haiku's standard maximum output-token limit.
- Knowledge cutoff
- August 2023 — assets.anthropic.com, read 16 October 2026
- API price, input
- $0.25 per 1M tokens — www.anthropic.com, read 16 October 2026
- API price, output
- $1.25 per 1M tokens — www.anthropic.com, read 16 October 2026
- Where
- API, Consumer app, Cloud marketplaces — www.anthropic.com, read 16 October 2026
- Parameters
- Not confirmed: The maker's pages reviewed do not publish the parameter count for Claude 3 Haiku.
- Model card
- Not confirmed: two readings disagree: https://assets.anthropic.com/m/61e7d27f8c8f5919/original/Claude-3-Model-Card.pdf and https://www.anthropic.com/claude-3-model-card
Benchmarks
Measured by others, not by the maker: each score with the name the measurer gives the model and the setting it ran, its day and its licence.
Arena
- Text Arena, claude-3-haiku-20240307: 1,262 (1,258–1,266, 117,701 votes), published 15 October 2026
- Vision Arena, claude-3-haiku-20240307: 1,000 (987–1,012, 13,380 votes), published 9 October 2026
Arena’s leaderboard dataset (lmarena-ai/leaderboard-dataset on Hugging Face), CC BY 4.0: ratings from people’s votes between two anonymous models, as Arena publishes them. Data · CC BY 4.0
Epoch AI
- Epoch Capabilities Index, Claude 3 Haiku: 118.4 (110.4–121.7), as of 19 October 2026
- GPQA Diamond, claude-3-haiku-20240307: 36.3% (± 2.4), run 27 January 2025
- OTIS Mock AIME 2024–2025, claude-3-haiku-20240307: 1.8% (± 0.8), run 25 February 2025
Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks, CC BY 4.0: Epoch AI’s own runs and its Capabilities Index. Data · CC BY 4.0
Questions
When was Claude 3 Haiku released?
Anthropic released Claude 3 Haiku on 13 March 2024.
What is the context window of Claude 3 Haiku?
The context window of Claude 3 Haiku is 200,000 tokens, as Anthropic prints it.
How much does the Claude 3 Haiku API cost?
On the standard tier, Claude 3 Haiku costs $0.25 per 1M tokens for input and $1.25 per 1M tokens for output, as Anthropic lists it.
Are the weights of Claude 3 Haiku open?
No: Anthropic does not publish the weights of Claude 3 Haiku.
What does Claude 3 Haiku take and return?
Claude 3 Haiku takes text, image and returns text.
What is the knowledge cutoff of Claude 3 Haiku?
The knowledge cutoff of Claude 3 Haiku is August 2023.
More from Anthropic
- Claude 1.0
- Claude 1.1
- Claude 1.2
- Claude 1.3
- Claude 2
- Claude 2.1
- Claude 3 Opus
- Claude 3 Sonnet
- Claude 3.5 Haiku
- Claude Fable 5
- Claude Fable 5.1
- Claude Haiku 4.5
- Claude Haiku 5.5
- Claude Instant 1.0
- Claude Instant 1.1
- Claude Instant 1.2
- Claude Mythos Preview
- Claude Opus 4.1
- Claude Opus 4.5
- Claude Opus 4.6
- Claude Opus 4.7
- Claude Opus 4.8
- Claude Opus 5
- Claude Opus 5.5
Sources
- Introducing the next generation of Claude (read 16 October 2026)
- Model deprecations (read 16 October 2026)
- Claude 3 Haiku: our fastest model yet (read 16 October 2026)
- The Claude 3 Model Family: Opus, Sonnet, Haiku (read 16 October 2026)
- Unsupervised Elicitation of Language Models (read 16 October 2026)