Google released Gemini 4 Neon on November 18, the middle of the Gemini 4 family: Carbon keeps the long-horizon projects, Hydrogen runs the fast execution loop, and Neon is aimed at the high-frequency work in between, complex coding, analysis and enterprise documents, at a lower cost per task than Carbon.

What it scores at the default high thinking level:

  • 61 on the Artificial Analysis Intelligence Index at xhigh, third overall, behind only the larger and slower models.
  • First on the Artificial Analysis Coding Agent Index at 58.6, with 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0.
  • A 1 million token context window, four thinking levels, $3 per million input tokens ($0.15 cached) and $15 per million output. Google puts the saving at 32% per task against Carbon at the default setting and 36% or more against competing models at their highest setting.

The cyber-tuned variant reaches 90.5% on SEC-bench Pro, 76% on the hard AutoBaxBench tier and 51.26% on WebExploitBench v1.2 at roughly half the cost per challenge of the larger frontier models, and goes to vetted defenders through the Fairwind Program in the coming days. Neon ships in the paid Gemini API, Google AI Studio, Gemini for business, and the Gemini and Antigravity apps for Pro and Ultra subscribers.

Related: Google Ships Gemini 4 Hydrogen as the Cheap Execution Model for Agents, Google Releases Gemini 4 Carbon, Google Releases Gemini 4 Argon

Introducing Gemini 4 Neon · Source channel