Anthropic launches Claude Haiku 5.5 for ultra cheap AI workloads

Haiku 5.5 is Anthropic’s clearest play yet to own the “cheap but smart” end of the model spectrum. By pricing input at around $0.10 per million tokens and output at $0.50, an order of magnitude below its previous small model, Anthropic is trying to make its stack the default choice for high volume classification, summarization, routing and sub-agent tasks that quietly consume most enterprise tokens. If those workloads consolidate around Claude, Anthropic gains both revenue and a powerful distribution channel into companies that may later upgrade to its larger frontier models.

Strategically, this release matters because it turns the Claude 5.5 line into a full family: Opus for peak capability, Sonnet for balanced work, and Haiku for scale. That mirrors NVIDIA’s good/better/best GPU strategy in software form. It also comes just as Anthropic heads toward a massive IPO, giving bankers and investors a clean story about product coverage across price points. Benchmark positioning against GPT-6 Luna is deliberate too; the message is that Anthropic can match or beat OpenAI on many practical metrics while being far cheaper to run at scale.

For the broader race to AGI, faster, cheaper small models push more tasks onto AI by default. That means more human work becomes either orchestrating agents or dealing with edge cases, accelerating both corporate dependence on these systems and the data flywheels that will train the next generation of frontier models.