The Information Machine
Updated todayNew4 sources

Google Launches Gemini 4 Neon, Ranks 3rd on Artificial Analysis Index

The gist

Neon combines frontier-level benchmark rankings with substantially lower cost per task than competing models, which is relevant for enterprises and developers choosing between high-capability models. Its cybersecurity-specific benchmark results extend its claimed applicability to penetration testing and code hardening workflows.

When stories like this move, I write about them in an email each morning.

One free email each morning, around 8 AM US Eastern.

The full picture

Google DeepMind released Gemini 4 Neon on November 18, 2026, positioning it between Gemini 4 Carbon and Hydrogen in its model lineup for complex software engineering, knowledge work, and scientific workflows. At its xhigh thinking level, Neon scores 61 on the Artificial Analysis Intelligence Index, placing it 3rd overall behind only larger, slower models, and 2nd on the Vals Index. It ranks 1st on the Artificial Analysis Coding Agent Index at 58.6, with its SWE-sweep score rising from 12.8% to 39.0% when run inside Google Antigravity. Neon also ranks 1st on Zapier's AutomationBench at 56.3% and on Vending-Bench 2 with $16,843, and 2nd on Agent Arena across more than 4,200 real agent sessions. On cybersecurity benchmarks, Neon scores 51.26% on WebExploitBench v1.2 at half the cost of competing large models, and DeepMind states it supports a full code hardening lifecycle covering vulnerability finding, validating, and patching. Pricing is $3 per million input tokens and $15 per million output tokens, with a 95% discount on cached inputs and a 1-million-token context window. DeepMind states the model delivers roughly 50-80% lower cost per task than top competing frontier models on the Vals Index and Agent Arena at comparable thinking levels, 32% less per task than Gemini 4 Carbon at its default high setting, and 36%+ less than competing models at their highest setting when using xhigh mode. On safety, Neon matched a 0.0% cheating rate on Terminal-Bench 4.0 and resisted all attacks in Gray Swan's indirect prompt injection benchmark at its highest thinking level.

How it developed
18 November 2026

Google DeepMind announced and began rolling out Gemini 4 Neon

Sources

Get the daily email

When stories like this move, I write about them in an email each morning.

One free email each morning, around 8 AM US Eastern.

See the latest edition · Privacy