Anthropic Releases Claude 3.5 Sonnet and Claude 3 Haiku Models

What Happened

Anthropic released two new large language models, Claude 3.5 Sonnet and Claude 3 Haiku, marking a significant expansion of the company's Claude 3 model family. The releases made both models available to developers and enterprise customers through Anthropic's API and its Claude.ai consumer platform.

The Models

Claude 3.5 Sonnet is positioned as Anthropic's most capable model at the time of release, sitting above Claude 3 Sonnet in the company's tiered lineup. Anthropic stated that Claude 3.5 Sonnet outperforms the prior Claude 3 Opus model on several benchmarks while operating at faster speeds and lower cost.

Claude 3 Haiku is the smallest and fastest model in the Claude 3 family. Anthropic described it as the most compact model in its class, designed for near-instant responsiveness and tasks that require high throughput at reduced cost. The company stated that Claude 3 Haiku can process a mix of image and text data and is optimised for enterprise workloads where speed and efficiency are priorities.

Benchmark Performance

According to figures released by Anthropic, Claude 3.5 Sonnet posted higher scores than Claude 3 Opus on multiple industry-standard evaluations including graduate-level reasoning tests, undergraduate-level knowledge assessments, and coding benchmarks. Claude 3 Haiku, while smaller, was benchmarked against competing models of similar size and price, with Anthropic reporting competitive performance on reasoning and language tasks relative to other compact models available at the time.

Pricing and Availability

Anthropic made both models available immediately through the Claude API. Claude 3.5 Sonnet was priced at three dollars per million input tokens and fifteen dollars per million output tokens. Claude 3 Haiku was priced at twenty-five cents per million input tokens and one dollar and twenty-five cents per million output tokens, making it the lowest-cost option in the Claude 3 family.

Both models support a context window of 200,000 tokens, consistent with the rest of the Claude 3 lineup. Vision capabilities, allowing the models to interpret and respond to image inputs alongside text, are included in both releases.

Background

Anthropic was founded in 2021 by former OpenAI researchers, including Dario Amodei and Daniela Amodei. The company has focused on AI safety research alongside its commercial model development. Anthropic has received substantial investment from Google and Amazon, with Amazon committing up to four billion dollars in a deal announced in 2023.

The Claude 3 model family, comprising Haiku, Sonnet, and Opus, was first introduced in March 2024. The three-tier structure was designed to give developers and enterprises a range of options balancing capability, speed, and cost. Claude 3 Opus was the flagship model at launch, while Haiku was positioned for high-volume, latency-sensitive applications.

What It Means in Practice

The release of Claude 3.5 Sonnet positioned Anthropic to compete directly with OpenAI's GPT-4o and Google's Gemini 1.5 Pro on performance benchmarks while offering pricing that Anthropic stated was lower than comparable models. Developers building applications that require a balance of high capability and cost efficiency gained an additional option in the competitive large language model market.

Claude 3 Haiku's pricing placed it in direct competition with smaller models from OpenAI and Google, targeting enterprise customers running large-scale inference workloads where per-token cost is a significant operational factor.

Anthropic indicated that further updates to the Claude model family, including additional versions in the Claude 3.5 series, were in development at the time of the release.