LLM API price tiers
Of the 55 models tracked, 10 cost under 0.25 per 1M input tokens, 10 cost 0.25 to 1, 18 cost 1 to 3 and 17 cost 3 or more. The cheapest single call at a 10,000-input / 1,000-output workload is Ministral 3 3B at $0.0011, and the most expensive is GPT-6 Cyber at $0.3000 (checked 2026-11-18).
The four price bands
10 models under 0.25 per 1M input tokens
| Model | Provider | Input /1M | Cached /1M | Output /1M | 1,000 calls (10K/1K) |
|---|---|---|---|---|---|
| Ministral 3 3B | Mistral | $0.1 | – | $0.1 | $0.0011 |
| Gemini 2.5 Flash-Lite | $0.1 | $0.01 | $0.4 | $0.0014 | |
| GPT-6 Luna | OpenAI | $0.1 | $0.01 | $0.5 | $0.0015 |
| Claude Haiku 5.5 | Anthropic | $0.1 | $0.01 | $0.5 | $0.0015 |
| Ministral 3 8B | Mistral | $0.15 | – | $0.15 | $0.00165 |
| GPT-4o mini | OpenAI | $0.15 | $0.075 | $0.6 | $0.0021 |
| Mistral Small 4 | Mistral | $0.15 | – | $0.6 | $0.0021 |
| Ministral 3 14B | Mistral | $0.2 | – | $0.2 | $0.0022 |
| GPT-5.6 Luna | OpenAI | $0.2 | $0.02 | $1.2 | $0.0032 |
| GPT-5.4 nano | OpenAI | $0.2 | $0.02 | $1.25 | $0.00325 |
10 models 0.25 to 1 per 1M input tokens
| Model | Provider | Input /1M | Cached /1M | Output /1M | 1,000 calls (10K/1K) |
|---|---|---|---|---|---|
| Codestral 25.08 | Mistral | $0.3 | – | $0.9 | $0.0039 |
| Gemini 3.1 Flash-Lite | $0.25 | $0.025 | $1.5 | $0.0040 | |
| DeepSeek Flash (V4.1) | DeepSeek | $0.3 | $0.006 | $1.2 | $0.0042 |
| GPT-5 mini | OpenAI | $0.25 | $0.025 | $2 | $0.0045 |
| Gemini 3.5 Flash-Lite | $0.3 | $0.03 | $2.5 | $0.0055 | |
| Gemini 2.5 Flash | $0.3 | $0.03 | $2.5 | $0.0055 | |
| Mistral Large 4 | Mistral | $0.7 | $0.07 | $2.1 | $0.0091 |
| Gemini 3.8 Flash | $0.75 | $0.075 | $3.75 | $0.0112 | |
| Gemini 3.6 Flash | $0.75 | $0.075 | $3.75 | $0.0112 | |
| GPT-5.4 mini | OpenAI | $0.75 | $0.075 | $4.5 | $0.0120 |
18 models 1 to 3 per 1M input tokens
| Model | Provider | Input /1M | Cached /1M | Output /1M | 1,000 calls (10K/1K) |
|---|---|---|---|---|---|
| Grok Build 0.1 | xAI | $1 | $0.2 | $2 | $0.0120 |
| GPT-6.1 Terra | OpenAI | $1 | $0.05 | $5 | $0.0150 |
| Claude Haiku 4.5 | Anthropic | $1 | $0.1 | $5 | $0.0150 |
| Grok 4.3 | xAI | $1.25 | $0.2 | $2.5 | $0.0150 |
| o4-mini | OpenAI | $1.1 | $0.275 | $4.4 | $0.0154 |
| DeepSeek V4 Pro | DeepSeek | $1.32 | $0.044 | $3.96 | $0.0172 |
| Gemini 2.5 Pro | $1.25 | $0.125 | $10 | $0.0225 | |
| Grok 4.7 | xAI | $2 | $0.5 | $6 | $0.0260 |
| GPT-4.1 | OpenAI | $2 | $0.5 | $8 | $0.0280 |
| GPT-6.1 Sol | OpenAI | $2 | $0.1 | $10 | $0.0300 |
| GPT-6 Sol | OpenAI | $2 | $0.2 | $10 | $0.0300 |
| Claude Sonnet 5.5 | Anthropic | $2 | $0.1 | $10 | $0.0300 |
| Claude Sonnet 5 | Anthropic | $2 | $0.2 | $10 | $0.0300 |
| Gemini 4 Carbon | $2 | $0.1 | $10 | $0.0300 | |
| Gemini 4 Argon | $2 | $0.1 | $10 | $0.0300 | |
| GPT-5.6 Terra | OpenAI | $2 | $0.2 | $12 | $0.0320 |
| Gemini 3.1 Pro Preview | $2 | $0.2 | $12 | $0.0320 | |
| GPT-5.4 | OpenAI | $2.5 | $0.25 | $15 | $0.0400 |
17 models 3 and over per 1M input tokens
| Model | Provider | Input /1M | Cached /1M | Output /1M | 1,000 calls (10K/1K) |
|---|---|---|---|---|---|
| Claude Sonnet 4.6 | Anthropic | $3 | $0.3 | $15 | $0.0450 |
| GPT-5.6 Sol | OpenAI | $4 | $0.4 | $20 | $0.0600 |
| Claude Opus 5.5 | Anthropic | $4 | $0.2 | $20 | $0.0600 |
| GPT-5.5 Rosalind Research | OpenAI | $5 | $0.5 | $25 | $0.0750 |
| GPT-5.5 Rosalind Discovery | OpenAI | $5 | $0.5 | $25 | $0.0750 |
| Claude Opus 5 | Anthropic | $5 | $0.5 | $25 | $0.0750 |
| GPT-5.5 | OpenAI | $5 | $0.5 | $30 | $0.0800 |
| Claude Fable 5.2 | Anthropic | $8 | $0.2 | $40 | $0.1200 |
| Claude Mythos 5.2 | Anthropic | $8 | $0.2 | $40 | $0.1200 |
| GPT-6 Astra | OpenAI | $10 | $1 | $50 | $0.1500 |
| GPT-6 Rosalind | OpenAI | $10 | $1 | $50 | $0.1500 |
| Claude Fable 5.1 | Anthropic | $10 | $0.25 | $50 | $0.1500 |
| Claude Mythos 5.1 | Anthropic | $10 | $0.25 | $50 | $0.1500 |
| Claude Fable 5 | Anthropic | $10 | $1 | $50 | $0.1500 |
| Claude Mythos 5 | Anthropic | $10 | $1 | $50 | $0.1500 |
| GPT-5.6 Cyber | OpenAI | $12.5 | $1.25 | $75 | $0.2000 |
| GPT-6 Cyber | OpenAI | $20 | $2 | $100 | $0.3000 |
How to read a price tier
A tier is the input rate per 1M tokens, not the cost of a call. A model at 0.1 per 1M input tokens still bills output tokens at its own output rate, and output is often several times the input price. Two models in the same tier can differ by an order of magnitude on cost per call once output length is taken into account, which is why the tables above also show the output rate and a 1,000-call cost for a fixed workload.
What a tier does not tell you
Cheap tokens are not cheap answers if the model needs retries, longer prompts or manual correction. Price tiers answer what a token costs, not what a finished task costs; for that, the benchmark comparison puts measured scores next to these rates and the coding ranking prices completed tasks. The cheapest-by-call ranking sorts the same models by a single fixed workload.
Related
- Every model price per 1M tokens
- Browse models by provider
- Cheapest LLM APIs, ranked by cost per call
- Blended cost by input:output mix
- LLM benchmark comparison
- Token cost calculator
- How prices are verified
Questions
What counts as a cheap LLM API?
On this site a cheap model is one whose standard input rate is under 0.25 per 1M tokens; 10 of the 55 models tracked fall in that band. Input rate is only half of the bill: check the output rate and the token shape of your workload before choosing.
Which models are in the lowest price band?
The lowest band (under 0.25 per 1M input tokens) contains 10 models; the full list and each model output rate are in the first table above. The band is defined by the input rate only, so a model with a low input rate and a high output rate can still be expensive for long answers.
Do cheaper price tiers mean lower quality?
Not by itself. A tier is a billing fact; quality has to be measured on your own prompts. This site does not publish a quality ranking of its own, but the benchmark comparison collects third-party scores with the source and read date so a price tier can be checked against measured results.
How is the 1,000-call cost in the tables calculated?
It uses the same formula as the calculator: 10,000 input tokens and 1,000 output tokens per call at the standard list rate, multiplied by 1,000 calls, with no caching and no batch discount. Change the token shape in the calculator to see how the order moves.
Next step → Browse models by provider
Price data last checked 2026-11-18 against official sources · how we verify
Price-change alerts
One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.