As predicted by AIM, Anthropic will officially roll out Claude Haiku 3.5 on Claude.ai. Haiku 3.5 is the latest version of Anthropic’s smallest and fastest model.
Although Haiku 3.5 is from the previous generation, it also surpasses Anthropic’s largest model, Claude 3 Opus. However, Haiku 3.5 doesn’t seem to support images as input yet, and Claude defaults back to Haiku 3 after uploading an image.
Haiku 3.5 is available on both Claude Web and mobile apps. The model was announced a month ago and is now available via Amazon Bedrock and Vertex AI’s API on Google Cloud. A few days ago, Anthropic also announced that they are optimizing Claude 3.5 Haiku on AWS Trainium 2.
“By enabling latency optimization, Claude 3.5 Haiku can accelerate inference speed by up to 60%, which is ideal for use cases ranging from code completion to real-time content moderation and chatbots. ,” Anthropic said in the announcement.
Anthropic also reduced the price of Claude 3.5 Haiku to $0.80 per million input tokens and $4 per million output tokens.
Additionally, Dario Amodei revealed a few weeks ago that he plans to release his flagship Opus 3.5 soon. Anthropic has apparently finished training Claude 3.5 Opus, which “scaled well and performed well,” according to the report.
The report also states that Anthropic did not release Claude 3.5 Opus because it used it for synthetic data generation and reward modeling that enhanced the functionality of 3.5 Sonnet.
The bell will ring. Mr. Amodei added, “We continue to scale up, and I have no doubt that we will come out with a model that is even more powerful than the one that currently exists.
“Or if we don’t, we’re making a major failure as a company,” he added.
But is Haiku 3.5 a little too slow? Google’s latest Gemini 2.0 Flash is a smaller model that delivers incredible performance, even better than the flagship Gemini 1.5 Pro. Forget about haiku. This model provides performance close to Claude 3.5 Sonnet in most benchmarks.
The new Gemini 2.0 Flash looks like a model monster.
It’s close to Claude 3.5 Sonnet in most of the benchmarks they show, and even better in MATH.
But at a price like 1.5 Flash, it’s essentially free by comparison.
It goes without saying that 2.0 Pro is coming… pic.twitter.com/kxNh6CJZWB
— Matt Schumer (@mattshumer_) December 11, 2024
Additionally, it provides multimodal capabilities including images, video, and audio, as well as multimodal output that combines natively generated images with text and steerable text-to-speech (TTS) multilingual audio.
