Cheapest AI models by capability tier
The least expensive model at each capability tier (1–8), by $/M output tokens, across 56 priced models in the Mintok catalog. The rule of thumb: pick the lowest tier that clears your quality bar, then take the cheapest model in it.
| Tier | Cheapest model | Vendor | Output $/M | Input $/M | Context |
|---|---|---|---|---|---|
| 8/8 | Claude Opus 4 | Anthropic | $75 | $15 | 200,000 |
| 7/8 | Gemini 3.1 Pro Preview | $12 | $2 | 1,048,576 | |
| 6/8 | DeepSeek V4 Pro | DeepSeek | $1.98 | $0.66 | 1,000,000 |
| 5/8 | GPT-5.6 Luna | OpenAI | $1.20 | $0.20 | 1,050,000 |
| 4/8 | LLaMA 4 Scout | Meta | $0.30 | $0.10 | 10,000,000 |
| 3/8 | Mistral NeMo | Mistral | $0.15 | $0.15 | — |
| 2/8 | Ministral 3 8B | Mistral | $0.15 | $0.15 | — |
| 1/8 | Ministral 3 3B | Mistral | $0.10 | $0.10 | — |
FAQ
What is the cheapest AI model?
The cheapest model in the Mintok catalog is Ministral 3 3B (Mistral) at $0.10 per million output tokens, at capability tier 1/8.
What is the cheapest capable model for hard tasks?
At the top capability tier (8/8), the cheapest option is Claude Opus 4 at $75 per million output tokens — the least expensive model that still clears that quality bar.
How should I pick the cheapest model that still works?
Pick the lowest capability tier that meets your quality bar, then take the cheapest model in that tier. Mintok’s Agent Cost Designer derives the required tier for a task and the Benchmark measures quality on your own prompts.
Find the cheapest model that actually works for you
The Agent Cost Designer derives the tier your task needs; the Benchmark measures quality on your own prompts.
Prices maintained by Mintok's catalog watchers from public sources. Capability tiers per methodology. Not investment advice.