Cheapest frontier AI models

The most capable tier, ranked by blended token price. The spread inside this group is over 25x, which means the choice of frontier model is a budget decision long before it is a capability one.

List prices checked

All 10, ranked

#NameProviderBlendedNotes
1DeepSeek-V4-ProDeepSeek$0.544/1M tokens1M context · $0.435 in / $0.870 out
2Mistral LargeMistral AI$3.00/1M tokens131K context · $2.00 in / $6.00 out
3GPT-5OpenAI$3.44/1M tokens400K context · $1.25 in / $10.00 out
4Gemini 2.5 ProGoogle$3.44/1M tokens1M context · $1.25 in / $10.00 out
5OpenAI o3OpenAI$3.50/1M tokens200K context · $2.00 in / $8.00 out
6Cohere Command ACohere$4.38/1M tokens256K context · $2.50 in / $10.00 out
7Claude Sonnet 5Anthropic$6.00/1M tokens200K context · $3.00 in / $15.00 out
8Grok 4xAI$6.00/1M tokens256K context · $3.00 in / $15.00 out
9Claude Opus 5Anthropic$10.00/1M tokens200K context · $5.00 in / $25.00 out
10Claude Fable 5Anthropic$20.00/1M tokens200K context · $10.00 in / $50.00 out

Source: DeepSeek (August 2026) · Mistral AI (May 2026) · OpenAI (May 2026) · Google (May 2026) · Cohere (May 2026) · Anthropic (August 2026). Published list prices only — committed-use discounts, regional variation, and per-request charges are not included. Not yet independently re-checked against the vendor’s page.

Questions

What is the cheapest frontier ai models in 2026?
DeepSeek-V4-Pro from DeepSeek tops this ranking at $0.544/1M tokens. The full table below shows all 10 entries with the same measure applied to each.
How is this ranking calculated?
The most capable tier, ranked by blended token price. The spread inside this group is over 25x, which means the choice of frontier model is a budget decision long before it is a capability one. Every figure is the provider's own published list price, linked in the source column and stamped with the month it was checked.
Why is the cheapest option not always the right one?
Because a single axis never captures the whole bill. A provider with the lowest storage rate may have the highest egress rate; a model with the cheapest input tokens may be the most expensive for agentic work. The comparison pages model both sides against identical workloads, which is the check worth doing before switching.

Other rankings