🎛️ Frugal AI Switchboard

Most cost-effective: ${models.reduce((a, b) => a.costNum a.costNum < b.costNum ? a : b).cost})
Greenest: ${models.reduce((a, b) => a.emissionsNum a.emissionsNum < b.emissionsNum ? a : b).emissions})
Most accurate: ${models.reduce((a, b) => a.arizeNum > b.arizeNum ? a : b).name} (${models.reduce((a, b) => a.arizeNum > b.arizeNum ? a : b).arizeAccuracy})
Most sovereign: ${models.reduce((a, b) => a.sovereignty > b.sovereignty ? a : b).name} (${models.reduce((a, b) => a.sovereignty > b.sovereignty ? a : b).sovereignty}%)

How to #QuitBigAI for smaller, local, open models… that are just as good!
💡 Coloured tabs = difference vs. baseline model (in grey, at the bottom)
On 📱? Horizontal mode ftw! 🙂

Select a category to view comparisons

Data sources: Energy & CO2: Ecologits; Accuracy: Arize Phoenix; Other data (pricing, parameters): Model cards

Frugal AI Switchboard was vibe coded by James Martin, BetterTech, on Mistral Vibe, which uses the Mistral Medium 3.5 model (128B). Vibe consumed c. 282k tokens to create this module. Re. Ecologits, this equates to 148Wh of electricity; 10.6 grammes of CO2 (emissions); 239 ml of water, and 235 µgSbeq of metal/mineral impacts.

Arize Phoenix is an open source platform for agent development and evaluation. Here, it provides data on models’ accuracy for the task in hand.

More on the making of the Frugal AI Switchboard in this blogpost.

Thanks to Tom Watson, RL Nabors, Marie Ait Daoud and more for the inspiration, and Florent Roques for the html help.

Keep your AI frugal; and ping me for any suggestions!

#QuitBigAI

Partager :

Like this: