Anthropic’s Claude 3.5 Haiku is Finally Here

As predicted by AIM, Anthropic is officially rolling out Claude Haiku 3.5 on Claude.ai. Haiku 3.5 is the latest iteration of Anthropic’s smallest and fastest model. 

Haiku 3.5 also surpasses Claude 3 Opus, which is Anthropic’s largest model, albeit from the previous generation. However, Haiku 3.5 doesn’t seem to support images as inputs yet, and Claude defaults back to Haiku 3 after uploading one.

Haiku 3.5 is available on both the Claude web and the mobile app. The model was announced a month ago and was made available via the API on Amazon Bedrock and Google Cloud’s Vertex AI.  A few days ago, Anthropic also announced that they’re optimising Claude 3.5 Haiku on AWS Trainium 2. 

“By enabling latency optimisation, Claude 3.5 Haiku can deliver up to 60% faster inference speed—making it the ideal choice for use cases ranging from code completions to real-time content moderation and chatbots,” said Anthropic in the announcement. 

Anthropic also lowered the price of Claude 3.5 Haiku to $0.80 per million input tokens and $4 per million output tokens. 

Moreover, Dario Amodei revealed a few weeks ago that there are plans to release the flagship Opus 3.5 Soon. As per reports, Anthropic has apparently finished training Claude 3.5 Opus, and ‘it performed well, with it scaling appropriately’. 

The report also said that Anthropic did not release it because it used Claude 3.5 Opus to generate synthetic data and for reward modelling to enhance the capabilities of 3.5 Sonnet. 

It rings a bell, doesn’t it? Amodei also said, “Scaling is continuing. There will definitely be more powerful models coming from us than the models that exist today. 

“Or if there aren’t, we’ve deeply failed as a company,” he added. 

But is Haiku 3.5 a tad bit too late? Google’s latest Gemini 2.0 Flash is a small model that offers tremendous performance, even better than the flagship Gemini 1.5 Pro. Forget Haiku; the model offers close performance with Claude 3.5 Sonnet on most benchmarks. 

Moreover, it also offers multimodal features, including images, video, and audio, as well as multimodal outputs such as natively generated images combined with text and steerable text-to-speech (TTS) multilingual audio.