Claude Haiku 4.5

Claude Haiku 4.5 (claude-haiku-4-5-20251001) is the bottom rung of Anthropic's price list and, in daily practice, the model that carries the highest request volume: classification, extraction, routing, moderation, first-line support — the work that is judged on cost per thousand calls rather than on benchmark scores. It costs 1 US dollar per million input tokens and 5 per million output, a fifth of what Sonnet 5 will cost once its introductory pricing ends on 31 August 2026, and Anthropic lists it as the fastest model in the family. It takes text and images and returns text, with a 200,000-token context window and up to 64,000 tokens of output. Extended thinking is supported, and this is where the model shows its age: Haiku 4.5 uses the older explicit switch, where the caller turns reasoning on for a request, while every model released after it — Sonnet 5, Opus 5, Fable 5 — moved to adaptive thinking, which decides on its own how much reasoning a question deserves. Haiku is the only current Anthropic model that does not have it. The rest of the line-up has also moved past it on context: Sonnet 5 and both Opus tiers now carry a million tokens, five times Haiku's window. Its training data runs to July 2025, with Anthropic marking February 2025 as the point up to which its knowledge is reliable — a year and a half behind the frontier models by mid-2026. None of that is an argument against it. On the tasks it is bought for, a 200,000-token window is not a constraint and a knowledge cutoff barely matters, because the relevant facts arrive in the prompt. What the gap does say is that Anthropic has not refreshed this tier since October 2025, while refreshing the tiers above it four times. For anyone planning a high-volume deployment, that is the number to keep an eye on: the cheap shelf is a year old, and a successor is overdue rather than announced.

▸Videos

No videos yet.

▸Reviews