DEV Community

News
News

Posted on

Google ships Gemini 4 Neon: frontier speed at $3 per million tokens

Infographic cover

๐Ÿ“… November 18, 2026 โ€ข Google Blog

Google released Gemini 4 Neon on November 18, a frontier-level model for fast end-to-end task completion in software engineering, knowledge work and science. Per the announcement from Koray Kavukcuoglu, SVP Google DeepMind, the new model sits between Carbon, built for heavy long-horizon projects, and Hydrogen, built for fast everyday execution. Per Latent Space and Arena, the lineup cadence ran 28 days from Argon to Carbon, then 15, then 6, making Neon the fourth lineup model in about seven weeks, with Argon retired.

Per the AI Studio docs, the model ID is gemini-4-neon, inputs are text, image, video, audio and PDF, and output is text only. The context window is 1,048,576 tokens, default max output is 65,536 tokens and long decode extends it up to 1M. Pricing is $3 per 1M input tokens, $15 per 1M output tokens and $0.15 per 1M cached, a 95% cache discount.

Google reports 61 on the Artificial Analysis Intelligence Index at xhigh, #3 overall, #2 on the Vals Index at 70.9%, a #2 Agent Arena debut at 1336, #1 on the Coding Agent Index at 58.6 and #1 on Zapier AutomationBench at 56.3%. Per Google, coding reaches 12.8% on SWE-sweep on the mini-SWE harness and 39.0% SOTA in Google Antigravity, with 99.2% agent honesty and a vendor-claimed 100% reflection of indirect prompt injection on Gray Swan IPI. The Decoder adds a caveat: #3 is not #1, since GPT-6.2 Astra and Claude Fable 5.2 rank higher and are absent from the Google table.

Speed per Artificial Analysis data cited by The Decoder is about 110 tokens per second versus 57 for Carbon, roughly 2x, cutting a benchmark run from 36 to about 21 minutes. The savings versus Carbon come from list price, not token efficiency, since Neon spends about 15% more output tokens at $6.44 versus $7.76 per task. A coding task in Antigravity comes to $6.86 per The Decoder and Artificial Analysis, versus $12.61 for Claude Opus 5.5 and $15.47 for Claude Fable 5.2. Until December 1 Carbon runs a promo at $2 and $10 per 1M, then $4 and $20, making Neon 50% more expensive against that promo.

The model is live as of November 18 in the Gemini API paid tier in Google AI Studio and Android Studio, in Gemini for business, in Antigravity for AI Pro and Ultra, and in the Gemini app for Pro and Ultra. Day-one integrations per Latent Space include Cursor, GitHub Copilot and LangChain, and per Simon Willison the llm-gemini 0.38 release added Neon to the llm CLI the same day. Per Google, confirmed by InfoWorld, Neon runs in Antigravity as an Orchestrator directing parallel Hydrogen subagents on multi-file debugging and refactoring.

Key figures

IMPACT for automation workflows: in an independent Activepieces test, Neon completed 3 of 3 tasks versus 2 of 3 for GPT-5.4 mini, but median latency was 6.4 seconds versus 1.0 and the per-task cost came out about 54x higher, near $8.30 versus $0.15 at 1,000 tasks per day. The Orchestrator pattern maps onto automation pipelines, with Neon planning while parallel Hydrogen subagents execute. The takeaway from the test is to verify price and latency on your own tasks before moving latency-sensitive or high-volume flows.

Sources:
blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon
aistudio.google.com/docs/models/gemini-4-neon
the-decoder.com/google-ships-gemini-4-neon-its-fourth-new-gemini-model-in-a-month
infoworld.com/article/4250618/google-launches-gemini-4-neon-touting-lower-costs-for-complex-ai-coding-tasks.html
activepieces.com/blog/gemini-4-neon-speed-accuracy-and-cost-tested
simonwillison.net/2026/Nov/18/llm-gemini
latent.space/p/ainews-gemini-4-neon-gdm-s-opus-sol-contender

IMPACT checklist


Sources:

  • blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon
  • aistudio.google.com/docs/models/gemini-4-neon
  • the-decoder.com/google-ships-gemini-4-neon-its-fourth-new-gemini-model-in-a-month
  • infoworld.com/article/4250618/google-launches-gemini-4-neon-touting-lower-costs-for-complex-ai-coding-tasks.html
  • activepieces.com/blog/gemini-4-neon-speed-accuracy-and-cost-tested
  • simonwillison.net/2026/Nov/18/llm-gemini
  • latent.space/p/ainews-gemini-4-neon-gdm-s-opus-sol-contender

Continue the conversation: https://t.me/AIAutomationArchitecture ยท Bluesky

AI Automation

Top comments (0)