Best Gemini model for agents and automations

For agents and automations with Gemini, use Carbon at medium effort.

Carbon at medium effort

Medium effortgeneration_config.thinking_level = medium

Scores 58 on the Artificial Analysis index at medium effort, above Argon's 50 at high, for less per task. Prices double on Dec 1. For long, multi-tool runs, use high.

How it compares

Prices are in US dollars per million tokens, input then output. Carbon costs $2 / $10.

ModelPriceAgainst the pick
Hydrogen$1.50 / $7.50Scores 41 at medium effort for about a sixth of Carbon's cost per task. Fine for simple, high-volume automations.
Argon$2 / $10Same token price. Scores 50 at high effort, below Carbon at medium (58).
3.8 FlashOlder$0.75 / $3.75Older Flash generation. Hydrogen replaces it for cheap automations.
3.5 Flash-Lite$0.30 / $2.50Fine for simple tool calls.

Setting it in the API

In an API request, set the model and effort like this:

model: "gemini-4-carbon", generation_config: { thinking_level: "medium" }

Maintained by the AI team at Vegan Hacktivists

Prices are per million tokens, input / output. We pick for the best result per dollar, so a pricier model can score higher and still lose out. This pick last changed Nov 13, 2026. Sources: Anthropic docs, OpenAI docs, Gemini docs, Artificial Analysis, Vals AI, Terminal-Bench.Last checked Nov 14, 2026.

Other picks

Gemini for code, everyday work, research.

Other labs for agents and automations: Claude, OpenAI.

Open the interactive picker to switch between labs and tasks.

Want helpputting this to work?

Vegan Hacktivists offers free AI and automation support to animal protection organizations — from a first conversation to a full build.

A beaver carrying a branch through the water