Loading...
Anthropic · OpenAI — Interactive pricing comparison
Haiku 4.5 is the more capable model—45 coding, full capability set including computer-use—while GPT-4.1 Mini is the budget floor at 42 coding with reasoning and computer-use stripped out. The question is whether Haiku's edge survives contact with its price.
Anthropic
OpenAI
GPT-4.1 Mini is 66% cheaper at this usage
GPT-4.1 Mini has a larger context window
These two look like neighbors until you do the output math. Input is close enough to ignore ($1 vs $0.4), but output tells the real story: Haiku charges $5 per million tokens against Mini's $1.6—a 3x spread. That ratio is the whole article. Haiku is genuinely the better model: it carries the full capability set, including computer-use that Mini simply doesn't have, and a few coding points on top. But capability is a fixed advantage while cost scales with every token you generate. In generation-dense workloads—batch content, high-fanout agents, anything that writes far more than it reads—Haiku's modest capability premium gets reabsorbed by that 3x output multiplier, and you end up paying triple for a 3-point edge you may not feel. Mini's smaller 200K-shy context (1M vs Haiku's 200K) actually flips here too: for long-context, output-light retrieval jobs, Mini is both cheaper and roomier. Haiku earns its price only when the work is light on output and actually needs computer-use or its capability headroom—a narrower band than the spec sheet suggests. Most high-volume generation pipelines will find the cheaper model is also the saner default.
Last updated June 2026