Digital strategy / Interactive cost model
AI model cost
with geopolitical risk.
Compare nine enterprise AI models on token cost, benchmark quality, and the chance that you must replace the model. Change the inputs, and see how the ranking changes.
Step 1 / Your workload
Change the inputs
The model starts with an example workload. Enter your own volumes and your own cost to replace a model.
Step 2 / The ranking
Cost for each quality point
Show the numbers as a table
| Model | Lab and country | Token cost | Chance (percent) | Risk cost | Risk-adjusted cost | Reasoning (HLE) | Coding (SciCode) | Cost per reasoning point |
|---|---|---|---|---|---|---|---|---|
| Mistral Large 4 | Mistral, France | $26,880 | 5 | $12,500 | $39,380 | 35.0 | 54.2 | $1,125 |
| Nemotron 3 Ultra | NVIDIA, USA | $25,920 | 5 | $12,500 | $38,420 | 28.4 | 40.3 | $1,353 |
| Muse Spark Open | Meta, USA | $50,400 | 10 | $25,000 | $75,400 | 43.9 | 58.4 | $1,718 |
| DeepSeek V4.1 Flash | DeepSeek, China | $12,960 | 40 | $100,000 | $112,960 | 43.3 | 55.9 | $2,609 |
| GLM-5.3 | Z.ai, China | $54,720 | 25 | $62,500 | $117,220 | 42.3 | 59.0 | $2,771 |
| Kimi K3.1 Pro | Moonshot, China | $96,000 | 25 | $62,500 | $158,500 | 52.5 | 60.8 | $3,019 |
| Claude Opus 5.5 | Anthropic, USA | $192,000 | 10 | $25,000 | $217,000 | 61.4 | 66.9 | $3,534 |
| Gemini 4 Argon | Google, USA | $192,000 | 10 | $25,000 | $217,000 | 57.1 | 61.8 | $3,800 |
| GPT-6 Astra | OpenAI, USA | $480,000 | 10 | $25,000 | $505,000 | 53.1 | 55.4 | $9,510 |
Step 3 / Your exposure
Set the chance of a forced change
These values are Devekko's judgment, not a measurement. Set each one from your customer contracts, your markets, and your license terms.
What can force a change
- Rules on origin
A government can bar a model for contracts or public systems. The rule also applies to a copy that you host.
- Customer contracts
A federal contractor or a regulated customer can require that you exclude a model. For that business, the chance is 100.
- Vendor decisions
The vendor of a closed model can change the price or the terms, or retire the model.
- Licenses
Open weights do not always mean an open license. Read each license before you depend on it.
How the model works
- 01
Token cost: input millions × input price, plus output millions × output price, times the months.
- 02
Risk cost: the chance of a forced change × the cost of the change.
- 03
Risk-adjusted cost: token cost plus risk cost.
- 04
Cost per point: risk-adjusted cost divided by the quality score.
The model does not include prompt caching, batch discounts, or router fees. A model below the minimum score moves to the end of the list.