Last month, Anthropic announced the updated Claude 3.5 Sonnet model and the new Claude 3.5 Haiku model. The new Claude 3.5 Sonnet model delivered modest improvements across the board, with significant gains in coding. Claude 3.5 Haiku, Anthropic"s answer to OpenAI"s GPT-4 Turbo Mini and Google"s Gemini 1.5 Flash, beats Claude 3 Opus, the largest model in Anthropic"s previous generation, on many AI benchmarks.
Anthropic announced today that Claude 3.5 Haiku is now available for developers via first-party API, Amazon Bedrock, and Google Cloud"s Vertex AI. For now, it is available only as a text-only model, and support for image input will be available in the future. To support developers who need image input, Claude 3 Haiku will remain available.
The most surprising part about Claude 3.5 Haiku is its pricing. Anthropic has priced Claude 3.5 Haiku at $1 per million input tokens and $5 per million output tokens, with up to 90% cost savings with prompt caching and 50% cost savings with the Message Batches API. At $1 per million input tokens, the new Claude 3.5 Haiku is nearly 7 times more expensive than OpenAI"s GPT-4o Mini and even more expensive than Google"s Gemini 1.5 Flash.
On X, Anthropic provided the following reason for such high pricing:
During final testing, Haiku surpassed Claude 3 Opus, our previous flagship model, on many benchmarks—at a fraction of the cost. As a result, we"ve increased pricing for Claude 3.5 Haiku to reflect its increase in intelligence.
Anthropic"s recommendation is to use Claude 3 Haiku for use cases that need image input or its lower price point.
With the ongoing price war between OpenAI and Google, it will be difficult for smaller competitors like Anthropic to survive. It will be interesting to see how Anthropic responds to developers" criticism around Claude 3.5 Haiku pricing. Anthropic may need to reconsider its pricing strategy to remain competitive in the rapidly evolving AI market. It"s crucial for them to strike a balance between profitability and accessibility to ensure the widespread use of their advanced language models.