Simple explanation
On October 7, 2026, Anthropic released Claude Haiku 5.5, its small, fast model. For requests under 100,000 tokens, which covers most everyday requests, the price is $0.10 per million input tokens and $0.50 per million output tokens. That is 90% less than Haiku 4.5. Larger requests get a smaller, 50% cut.
Those prices match OpenAI's GPT-6 Luna, the model behind ChatGPT's Free plan. So both big AI companies now offer a small model at the same low price.
Why this matters if your app uses AI: many apps send every request to the biggest model available, because that is what the tutorial used. But a lot of AI work in a real app is simple:
- tagging a message as a question, complaint or order
- writing a two-line summary
- checking whether a form answer looks valid
- turning text into a fixed format
Small models are built for exactly these jobs. Big models are still worth paying for when the task needs careful reasoning, long writing or tricky judgement. The skill is knowing which job is which.
Note that these are the companies' own prices and performance claims. Test on your own tasks before you switch.
What to do
Sort your app's AI tasks before you change anything:
1. List every place your app uses AI, with one line describing each job.2. Mark each job SIMPLE (sorting, short summaries, yes or no checks, formatting) or HARD (reasoning, long writing, advice, anything a customer reads as the final answer).3. Pick one SIMPLE job to test first. Choose the one your app runs most often, because that is where the savings are.4. Run your 5 saved test prompts for that job on the small model. If you do not have a test set yet, make one first; the model-change post linked on this page shows how.5. Switch that one job only if every test passes. Keep a note of the old model name so you can switch back.6. Leave HARD jobs on the bigger model. Saving a few cents is not worth a worse answer for your customer.
Repeat for the next simple job a week later, once you have seen the first switch working with real users.
Copy-paste prompt
Here are the places my app uses AI: [LIST EACH FEATURE AND WHAT THE AI DOES]. I am not a developer. For each one, tell me: 1. is this a SIMPLE job a small, cheap model can usually handle, or a HARD job that needs a bigger model, and why; 2. what could go wrong if a small model handles it; 3. one test I can run to check. Then tell me which single job to test on a small model first. Do not change any code; this is planning only.
Course note
Key takeaway
Small models are now very cheap. Move your simple, high-volume AI jobs to them one at a time, test each switch, and keep the hard jobs on a bigger model.