Claude Haiku 4.5 is the model Magnus ships with by default: quick, cheap, and far more capable than a "small" model has any right to be. For voice conversations, its low latency is the difference between talking to software and talking to someone.
Launched October 15, 2025 by Anthropic. The pitch: performance comparable to Claude Sonnet 4 - the previous mid-tier flagship - at one third of the cost and more than twice the speed. API pricing is $1 per million input tokens and $5 per million output.
On the All-inclusive plan a Haiku 4.5 reply costs 1 credit - the most conversation per credit of any Claude model.
Everyday back-and-forth, voice mode, quick-fire agents, and high-volume days where credits matter. It still runs the full Claude toolset - integrations, web search, PDFs - so "fast" does not mean "stripped down".