Introducing Gemini 4 Neon
2026-11-20 · Google DeepMind
Introducing Gemini 4 Neon
On November 18, 2026, Google DeepMind introduced Gemini 4 Neon, described by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google, as bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.
Model position, pricing, and availability
- At its xhigh thinking level, Neon scores 61 on the Artificial Analysis Intelligence Index and ranks #3 overall, behind larger, slower models. It ranks #2 on the Vals Index and lands on the Pareto frontier. On the new Agent Arena Index, it ranks #2 for steerability—adapting cleanly to mid-task user corrections.
- Neon also excels at multi-step instruction following and records Google’s lowest fabrication rate yet in internal Google Antigravity testing.
- Pricing is $3 per million input tokens, $0.15 per million cached input tokens (a 95% discount), and $15 per million output tokens, with a 1 million-token context window.
- It is rolling out today in the Gemini API in Google AI Studio, Gemini for business via API and model dropdown, for paid Google AI Pro and Ultra subscribers in Google Antigravity, and for Pro and Ultra subscribers in the Gemini app.
Enterprise, analytical, and scientific work
Neon can synthesize spreadsheets, PDFs, charts, audio, and video in a single 1 million-token context window and turn inputs into structured deliverables. Independent evaluations rank it #2 on the Vals Index (70.9%) and #1 on both the Legal Research Bench and Tax Agent Bench. On Surge AI, it ranks #1 on most benchmarks across finance, accounting, spreadsheet modeling, and complex constraint following. It scores 73.5% on Chartography, 75.8% on Agents’ Last Exam v1.1 (on par with leading models), #1 on Zapier’s AutomationBench (56.3%), and #1 on Andon Labs’ Vending-Bench 2 ($16,843), a long-horizon business simulation. It debuts #2 globally with 1336 on Arena’s new Agent Arena Index across 4,200+ real-world agent sessions. For scientific and mathematical reasoning, it is best in class on Terminal-Bench Science (70.5%), RiemannBench (84.0%), and HLE-Diamond (71.3%).
Production-ready code
Engineered for long-running agent workflows, Neon sustains performance through complex, tool-heavy tasks end to end, whether running standalone or as a fast, high-judgment Orchestrator directing parallel Hydrogen subagents in Google Antigravity. It cuts redundant file reads, shares progress, and handles context compaction reliably in long sessions. It ranks #1 overall on the Artificial Analysis Coding Agent Index (58.6), scores 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default high setting. For production code, it handles multi-file debugging and refactoring across repositories with targeted, surgical edits. On SWE-sweep, it leads on the mini-SWE harness (12.8%) and more than triples to a state-of-the-art 39.0% in Google Antigravity.
Examples and efficiency
In Antigravity, Neon used /swarm with Blender MCP to build a photorealistic 3D lab bench visualizing model names with elemental accuracy; wrote a first-person WebGL2 freeride ski simulator from scratch on real Mont Blanc and Chamonix terrain, with 10,200 lines of custom shaders, 120 Hz ski physics, six alpine lines, dynamic carved tracks, slab avalanches, and spatial Web Audio with zero pre-made assets; recreated Alan Turing’s 1936 study at King’s College, Cambridge in interactive 3D, including a working Universal Turing Machine; and turned a single image into a 100,000-cell Google Sheet with zero embedded images and zero Apps Script, using over 500,000 live formulas for edge detection, 3D depth mapping, and relighting across six interactive modes.
Neon has four thinking levels that scale steadily. At default high, it pairs a high Artificial Analysis Intelligence Index score with 32% lower cost per task than Carbon. It sits on the Pareto frontier on the Vals Index and Agent Arena at high, with roughly 50% to 80% lower cost per task than top competing frontier models. xhigh adds deeper, multi-step reasoning for the hardest engineering, analytical, and scientific workloads and shows its highest intelligence score.