News
November 19, 2026 · Google

Google adds Gemini 4 Neon, a mid-priced model for coding and agent work

Google released Gemini 4 Neon on November 18, charging $3 for every million input tokens and $15 for every million output tokens. It reads up to 1 million tokens of context.

Google introduced Gemini 4 Neon on November 18, 2026, in a post by Google DeepMind's Koray Kavukcuoglu. The company pitches it as a fast model for software engineering, office work and science tasks. According to The Decoder, Neon is the fourth Gemini 4 model Google has released in about four weeks.

What's new

  • Price: Google charges $3 for every million input tokens and $15 for every million output tokens. Cached input costs $0.15 per million, which Google describes as a 95% discount.
  • Context and inputs: the model has a 1 million-token context window. The Decoder adds that Neon accepts text, images, video, audio and PDFs, but cannot generate images or audio.
  • Benchmarks (Google's figures): Google says Neon scores 61 on the Artificial Analysis Intelligence Index at its top "xhigh" thinking level, which puts it third overall, and that it ranks second on the Vals Index.
  • Thinking levels: Neon has four levels. At the default "high" setting, Google claims a 32% lower cost per task than Gemini 4 Carbon.
  • Agent work: Google says Neon can act as an orchestrator that directs parallel Hydrogen subagents inside its Antigravity coding tool.

Availability and pricing

Google says Neon is rolling out in the Gemini API through Google AI Studio, in Gemini for business, in Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. For comparison, The Decoder lists Gemini 4 Hydrogen at $1.50 per million input tokens and $7.50 per million output tokens. Carbon and Argon are at an introductory $2 and $10, rising to $4 and $20 on December 1.

Why it matters

Analysis: once the introductory prices for Carbon and Argon end, Neon will sit between Google's cheapest Gemini 4 model and its two larger ones. Developers picking a default model for coding agents now have another Google option to test against GPT-6.2 and Claude. The benchmark results above are Google's own, so independent evaluations are worth watching before switching.

Source: Google