We noticed one number: Anthropic cut Claude Haiku 5.5's API pricing from the previous generation's $1/$5 (per million tokens input/output) straight to $0.10/$0.50—a 90% drop, landing in lockstep with OpenAI's GPT-6 Luna released last month. This isn't a routine upgrade; it's a signal that the small-model API price war has broken out in plain sight.
What This Is
Haiku is Anthropic's lightweight model line, built for speed and low cost. Haiku 4.5, the year-old predecessor, sat at $1/$5—already on the pricey side, ten times what GPT-6 Luna cost.
Haiku 5.5 slashes prices to $0.10/$0.50—identical to Luna within 100K tokens (a token is the model's billing unit, roughly 0.7 English words or 1.5 Chinese characters); above 100K, pricing jumps 5x to $0.50/$2.50. Sonnet 5.5's cache reads (cached reads, the discounted billing for reusing already-processed content) are halved at the same time.
The subscription side moves harder: Max 5x users get $100 in monthly API credits, Max 20x gets $200, and Team plans pool up to $500; credits don't roll over—use it or lose it.
Industry View
Within 100K tokens, Haiku 5.5 benchmarks above Luna at the same price tier; API credits can be applied, and auto-recharge can be disabled to keep bills from spiraling—friendly signals for developers.
But there are several warning signs worth flagging. First, above 100K tokens Haiku jumps 5x, while Luna only ticks up slightly to $0.20/$0.75 at 272K tokens—for long-context scenarios like lengthy reports or full-book analysis, Luna is actually the better deal. Second, the new tokenizer (the tool that splits text into billable units) has gotten "stingier": the same text consumes roughly 25% more tokens than the previous generation—classic "discount headline, hidden markup" tactics. Third, subscription credits expire if unused, which is a soft lock-in for light users.
Impact on Regular People
For enterprise IT: batch processing tasks within 100K tokens (customer service, document summarization, content moderation) can drop to one-tenth of previous per-call costs; but long documents, lengthy meeting minutes, full codebases, and similar scenarios need fresh math.
For working professionals: the marginal cost of using AI to draft emails, organize materials, and translate drops further; subscribers now have API credits layered on top, making "API + chat" hybrid usage the new normal.
For the consumer market: AI subscription pricing still has room to fall, or we'll see more "monthly subscription + usage-based" hybrid products—the savings will eventually show up in final pricing.