Cheapest Frontier AI Models
Ranked from provider-published pricing · Prices checked 12 August 2026
Frontier models are the most capable tier each lab ships. They are the expensive end of the market, so the spread between them matters: this list ranks only frontier-class models by blended cost per million tokens.
The ranking
top 20 of 41Llama 4 Maverick from Meta leads this ranking at $0.320/M. That is 92% below the $4.00/M median across the 41 models that qualify for this list.
| # | Model | Provider | Top-Tier Models Ranked by Blended Cost | Input /1M | Output /1M | Context |
|---|---|---|---|---|---|---|
| 1 | Llama 4 Maverick | Meta | $0.320/M | $0.200 | $0.600 | 128,000 |
| 2 | DeepSeek-V3 | DeepSeek | $0.519/M | $0.270 | $1.10 | 64,000 |
| 3 | DeepSeek-V3-0324 | DeepSeek | $0.519/M | $0.270 | $1.10 | 128,000 |
| 4 | Mistral Large 3 | Mistral | $0.800/M | $0.500 | $1.50 | 256,000 |
| 5 | Doubao-Pro-128k | ByteDance | $0.812/M | $0.560 | $1.40 | 128,000 |
| 6 | Palmyra X 004 | Writer | $0.950/M | $0.500 | $2.00 | 128,000 |
| 7 | MiniMax-Text-01 | MiniMax | $1.33/M | $0.700 | $2.80 | 1,000,000 |
| 8 | Nova 2 Pro | Amazon | $1.52/M | $0.800 | $3.20 | 1,000,000 |
| 9 | Nova Pro | Amazon | $1.52/M | $0.800 | $3.20 | 300,000 |
| 10 | Grok 4.3 | xAI | $1.63/M | $1.25 | $2.50 | 1,000,000 |
| 11 | Muse Spark 1.2 | Meta | $2.15/M | $1.25 | $4.25 | 1,049,000 |
| 12 | DeepSeek-V4-Pro | DeepSeek | $2.26/M | $1.74 | $3.48 | 1,000,000 |
| 13 | Solar Pro | Upstage | $2.85/M | $1.50 | $6.00 | 32,000 |
| 14 | Yi-Large | 01.AI | $3.00/M | $3.00 | $3.00 | 32,000 |
| 15 | Qwen2.5-Max | Qwen | $3.04/M | $1.60 | $6.40 | 32,000 |
| 16 | Mistral Large 2 | Mistral | $3.20/M | $2.00 | $6.00 | 128,000 |
| 17 | GPT-4.1 | OpenAI | $3.80/M | $2.00 | $8.00 | 1,000,000 |
| 18 | Jamba 1.5 Large | AI21 Labs | $3.80/M | $2.00 | $8.00 | 256,000 |
| 19 | GPT-5 | OpenAI | $3.88/M | $1.25 | $10.00 | 400,000 |
| 20 | Gemini 2.5 Pro | $3.88/M | $1.25 | $10.00 | 1,000,000 |
How this list is built
- Ranked cheapest first by a blended rate weighting input and output at 70/30. Default reference rate. 70% weight on input, 30% on output.
- Excludes embed models.
- Includes models marked available or preview; retired and deprecated SKUs are excluded.
- Prices are the rates each provider publishes, not estimates. How the blended rate is calculated.
Frequently asked questions
What counts as a frontier model?
For this list, a model our catalogue tags as its provider’s top-of-line tier (flagship or frontier-class) rather than the broader family of capable models. It is a capability tier, not a price bracket, though frontier models are usually the priciest option a provider sells.
Related rankings
Price it for your workload
Rankings are recomputed on every deploy from the catalogue. Prices checked 12 August 2026.