LLM pricing · Compare
In short: Claude Haiku 5.5 is 10× cheaper than Claude Haiku 4.5 per chat message: $0.00035 vs $0.0035 for a typical chat turn (1,500 tokens in, 400 out). On a $9.99 plan, a user can send about 24,261 messages a month on Claude Haiku 5.5 before they cost more than they pay, versus 2,426 on Claude Haiku 4.5.
| Claude Haiku 4.5 | Claude Haiku 5.5 | |
|---|---|---|
| Provider | Anthropic | Anthropic |
| Input | $1 | $0.1 |
| Cached input | $0.1 | $0.01 |
| Output | $5 | $0.5 |
| Long prompts | same rate | over 100,000 tokens: $0.5 in, $2.5 out |
| Claude Haiku 4.5 | Claude Haiku 5.5 | |
|---|---|---|
| Per message | $0.0035 | $0.00035 |
| Per message, prompt 2/3 cached | $0.0026 | $0.00026 |
| Light user (30 msgs/mo) | $0.105 | $0.0105 |
| Typical user (300 msgs/mo) | $1.05 | $0.105 |
| Heavy user (3,000 msgs/mo) | $10.50 | $1.05 |
| Break-even on $9.99 | 2,426 msgs/mo | 24,261 msgs/mo |
Red: the user costs more in AI than the $8.49 you keep from a $9.99 plan after a 15% store fee.
Price is only half of it: pick the cheapest model that gives answers your users accept. Many apps route most traffic to the cheaper model and send only hard requests, or only paying users, to the stronger one. Whatever the mix, the number that decides your margin is cost per user, because a few heavy users usually drive most of the bill.
List prices, standard tier, prompts under the long-context threshold. Averages hide your heaviest users: FiscaLLM shows this per real user.
FiscaLLM prices every Anthropic call with these rates, per user, and sets it next to what that user paid through RevenueCat, Stripe or a CSV. You see which users and which model lose you money. Free and open source. See the demo.
Claude Haiku 5.5 is 10× cheaper than Claude Haiku 4.5 per chat message: $0.00035 vs $0.0035 for a 1,500-token prompt and a 400-token reply, at list prices.
About 2,426 a month on Claude Haiku 4.5 and 24,261 on Claude Haiku 5.5, before AI cost uses up what you keep after a 15% store fee.
Yes. A common pattern is the cheaper model for most traffic and the stronger one for hard requests or paying users. Track cost per user to see whether the mix pays off.
List prices, standard tier. Batch, priority and regional pricing differ; check each provider for the latest.