Anthropic Cuts Prices for Claude 3.5

Anthropic has reduced the cache read pricing for its latest model, Claude 3.5 Haiku. The company also explains that this model offers performance equivalent to Claude 3.5 Sonnet while differing in its safety-related settings.

Price competition in the AI model market has intensified rapidly over the past several years. As major players like OpenAI and Google have successively announced cost reduction measures, Anthropic finds itself compelled to respond competitively. Cache reading is a mechanism that lowers API call costs by reusing data that has already been processed, and for companies and developers that frequently invoke models, it represents a critical billing item directly tied to actual expenditures.

Anthropic's price adjustment this time is limited to the cache read component. According to the company, Claude 3.5 Haiku is said to possess capabilities close to those of Claude 3.5 Sonnet, but there are differences in safety-related settings (safeguards) between the two. As of now, detailed information about the specific magnitude of changes or the nature of safety differences has not been confirmed.

The fact that the price reduction is limited to a specific billing item merits attention. Rather than a general price cut, this targeted adjustment to the frequently-used cache read function can be positioned as a strategic decision aimed at capturing cost-conscious enterprise users. For companies integrating AI models into their operational systems, API running costs are a major consideration in model selection, and some observers note that even partial price reductions have non-trivial practical impact.

At the same time, the point that two models claimed to have equivalent performance carry different safeguards is an element that users cannot ignore. An AI model's safety settings directly affect output control and handling of harmful content. Depending on use cases, model selection criteria may extend beyond cost to include safety design considerations, making how Anthropic explains and discloses these differences a key factor for future decisions.

Across the AI industry overall, a trend of simultaneous model performance improvement and cost reduction is becoming established. Anthropic's current move can be understood as part of this broader trend. As users, the perspective of selecting models based not solely on simple price comparison but also on differences in safety design and usage conditions is increasingly required.