Gemini 4 Neon is "#1 overall" for coding agents, says Google's Koray Kavukcuoglu

Google released Gemini 4 Neon on 18 November 2026. Its launch post ranks the model first on the Artificial Analysis Coding Agent Index and third on its Intelligence Index; the benchmark's own pages show a lead of 0.1 points in coding and a sixth place among 95 entries.

An orange neon sign reading Gemini 4 Neon on a dark wall, the header picture of Google's launch post
Picture: Google
In this material
  1. What Artificial Analysis shows
  2. Context

Google published Gemini 4 Neon on 18 November in the Gemini API, Google AI Studio, Gemini for business and, for paid AI Pro and Ultra subscribers, the Gemini app and the Antigravity coding tool. The launch post is signed by Koray Kavukcuoglu, senior vice president of Google DeepMind and Google’s Chief AI Architect, and it puts coding first.

Across core developer and researcher workflows, Neon achieves frontier performance and is competitive with the leading models in its class, ranking #1 overall on the Artificial Analysis Coding Agent Index (58.6), with a score of 65.2% on Terminal-Bench 4.1 and 46.2% (at its default high setting) on FrontierCode 2.0.

Koray KavukcuogluSenior vice president, Google DeepMind, and Chief AI Architect, GoogleGoogle blog, November 18, 2026

The price is $3 per million input tokens, $0.15 for cached input and $15 per million output tokens. Developers choose among four thinking levels, from low to xhigh; high is the default.

What Artificial Analysis shows

Artificial Analysis is an independent benchmarking company. Its Coding Agent Index scores a coding tool and a model together on three test sets: FrontierCode-AA 2.0, Terminal-Bench 4.1 and SWE-Atlas-QnA-AA. On 19 November its table read:

Antigravity CLI, Gemini 4 Neon (Xhigh)
58.6
Claude Code, Claude Fable 5.2 (Xhigh)
58.5
Muse Code, Muse Wonder Full (Max)
54.5
Codex, GPT-6.2 Astra (Max)
53.5

The same page puts Neon’s average cost at $6.86 per task.

The post makes a second ranking claim:

Neon ranks ahead of comparable models and #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, landing just behind larger, slower models.

Koray KavukcuogluSenior vice president, Google DeepMind, and Chief AI Architect, GoogleGoogle blog, November 18, 2026

The Intelligence Index leaderboard lists Claude Fable 5.2 at 65, GPT-6.2 Astra at 62 and Gemini 4 Neon at 61. The page Artificial Analysis keeps for the model gives Neon “#6 / 95” in intelligence. It calls the model “somewhat expensive when comparing to other models of similar price” and slow per task, at $6.44 per task of the index. AI Weekly built its report on that page: Artificial Analysis “places Neon at #6, not #3”.

Context

Neon follows Gemini 4 Carbon, released on 28 October, and the post compares the two: at the default high level, it says, Neon costs 32% less per task. OpenAI released GPT-6.2 Astra a day before Neon. Artificial Analysis adds new results as it tests more tools and settings, so its tables may change.

From the pulse

All news →
  1. Models

    Google replaces Gemini 3.8 Flash with Gemini 4 Hydrogen across its products

    Google says Hydrogen is 2.5 times faster than 3.8 Flash even at its high thinking level and makes more than five times fewer fabricated claims. Unlike the earlier Gemini 4 Argon and Carbon, it was opened to users at once, and Search got an upgrade built on it the same day.

    Sources: arstechnica.com, blog.google In the catalog: Gemini

  2. Hardware

    Smart glasses with Gemini from Warby Parker and Gentle Monster go on preorder from $395

    Samsung makes the hardware and Google the software. Warby Parker sells three frames from $395 in the US only; Gentle Monster's six styles start at $449 in the US and South Korea.

    Source: arstechnica.com In the catalog: Gemini

  3. Models

    Gemini Omni 1.2 Flash makes 15-second videos and costs 30% less

    Google's video model now generates and edits clips of up to 15 seconds natively, keeps scenes more consistent and is 30% cheaper than before.

    Source: blog.google In the catalog: Gemini

More in this section

All materials →
News

Shinji Mikami wishes AI would "all just go away"; Kim Hyung-tae says games can't avoid it

At G-CON 2026 in Busan on 19 November, the director of the first Resident Evil and the head of Shift Up described how their studios use AI. Mikami keeps it away from design; Kim said game development without it no longer exists.

News

Cliff Bleszinski on Big Walk's AI backlash: "People really need to get over themselves"

House House said an AI chatbot helped with about 10 lines of Big Walk's code after IGF jurors refused to vote for games made with generative AI. Negative Steam reviews followed, and the Gears of War designer took the studio's side.

News

OpenAI ships GPT-6.2 Astra, its first public model rated High for self-improvement

OpenAI released GPT-6.2 Astra and Sol on 17 November 2026. Its system card makes Astra the first model the company has released that meets its own High threshold for AI self-improvement; Epoch AI's test of real work tasks put it at 65%.