Anthropic: Claude Haiku 5.5

Claude Haiku 5.5 is a text model from Anthropic, and it is the successor to Claude Haiku 4.5. Anthropic describes it as its small, fast model, aimed at high volume and cost sensitive work. If you have used earlier Haiku releases, this one is the direct replacement in that line, with the improvements Anthropic lists over 4.5 being stronger coding, computer use, and related agentic behaviour.

On the input side the model accepts text, images, and files, and it produces text. The context window is 1,000,000 tokens, which is large enough to hold long documents, codebases, or extended conversation histories in a single request. The maximum output is 128,000 tokens, so the model can produce lengthy responses rather than being capped at short completions. Those two numbers together mean you can send a substantial amount of material and still ask for a substantial amount back.

The provider description points at the workloads this model is built for: summarization, subagents, and browser use, all at high volume. Summarization fits naturally because you can pass large documents or file collections in the context window and get condensed text out. Subagents fit because a fast model can sit behind a planner, handling the smaller tool calls and intermediate steps that would be wasteful to route to a larger model. Browser use and computer use applications, along with coding tasks, are areas where Anthropic says this release improves on 4.5, so agent style loops that inspect pages or read and write code are a reasonable fit.

You call it through api14's OpenAI compatible API using the model id "anthropic/claude-haiku-5.5", pointing your existing OpenAI client at the gateway. The request parameters it supports are include_reasoning, max_completion_tokens, max_tokens, reasoning, reasoning_effort, response_format, stop, structured_outputs, tool_choice, tools, and verbosity. In practice that means you can control output length with max_tokens or max_completion_tokens, constrain the response shape with response_format or structured_outputs, define functions with tools and steer selection with tool_choice, apply stop sequences, and tune how much reasoning the model surfaces or performs through reasoning, reasoning_effort, include_reasoning, and verbosity.

Because this is positioned as Anthropic's small and fast model, it is the sensible default when throughput and volume matter more than depth on a hard problem. For tasks that need the strongest possible reasoning, a larger model in the same family may be the better choice, and you can keep the same integration since the request shape is compatible. The 1,000,000 token context and 128,000 token output ceiling do remove some of the usual reasons to reach for something bigger, so check whether the smaller model already meets your needs before switching.