On 19 October 2026, public list prices for small models start at about $0.10 per million input tokens, mid tier models cluster around $2 input and $10 to $12 output, and large models list from $4 to $10 input. Cached input is far cheaper. Prices below are standard processing and will change.
How to read this
All figures are US dollars per million tokens, taken from each provider's public pricing page on 19 October 2026. The groupings are mine, based on how each provider positions its models, not an official ranking. Models are not equivalent, so this table is a price map, not a quality comparison.
The tables
Small models
| Model | Input | Cached input | Output |
|---|---|---|---|
| Anthropic Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 | $0.01 | $0.50 |
| OpenAI gpt-6-luna | $0.10 | $0.01 | $0.50 |
| Google Gemini 3.1 Flash-Lite | $0.25 | $0.025 | $1.50 |
Mid tier models
| Model | Input | Cached input | Output |
|---|---|---|---|
| Anthropic Sonnet 5.5 | $2 | $0.10 | $10 |
| OpenAI gpt-6.1-sol | $2.00 | $0.10 | $10.00 |
| Google Gemini 3.1 Pro Preview (up to 200k) | $2.00 | $0.20 | $12.00 |
Large models
| Model | Input | Cached input | Output |
|---|---|---|---|
| Anthropic Opus 5.5 | $4 | $0.20 | $20 |
| Anthropic Fable 5.1 | $10 | $0.25 | $50 |
| OpenAI gpt-6-astra | $10.00 | $1.00 | $50.00 |
| Google Gemini 4 Argon | $2.00 (scheduled to rise to $4.00 on 1 December 2026) | $0.10 | $10.00 (scheduled to rise to $20.00) |
What the table does not show
- Batch or flex pricing, which lists about half of standard on all three providers. See batch API discounts.
- Long context rates. See long context pricing.
- Cache write charges, which on Anthropic's page are 1.25 times base input for five minutes, and Google's storage fees per million tokens per hour.
- Tool charges, such as search grounding or web search.
- Regional processing uplifts. Anthropic lists a 1.1 times multiplier for US only inference on newer models, and OpenAI's page lists a 10 percent uplift for regional processing on eligible models.
- Tokenizer differences. The same text can be a different number of tokens on different models, so a list price per token is not a price per document.
Three things the snapshot suggests
- The price spread between small and large models on one provider is large. On Anthropic's page, Haiku 5.5 input is 40 times cheaper than Opus 5.5 input. That makes routing worthwhile. See choose a model tier.
- Output costs five or six times input in every row, so output length matters.
- Cached input is 10 to 20 times cheaper than fresh input in most rows, so stable prefixes are worth designing for.
How long will this stay true
Not long. Google's page already lists scheduled increases, and Anthropic's page shows a cancelled one. See scheduled price changes in AI APIs. Use this post to learn the shape, and the source pages to price anything real.
Quick check
Which provider is cheapest?
That depends on the task, the model you need and your cache hit rate. Compare cost per finished task on your own examples.
Why are some models missing?
This is a short snapshot, not a full catalog. Each provider lists more models, including older ones.
Sources
Last reviewed: 19 October 2026.
