Google launches Gemini 4 Neon on Gemini, Antigravity, and APIs
Google launched Gemini 4 Neon, a mid-tier model with a 1M-token context window, outperforming Claude Opus 5.5 on CursorBench at lower cost.
Decoder
- CursorBench: A coding-specific benchmark used to evaluate how effectively AI models can complete software engineering tasks within an IDE.
- Context window: The maximum amount of information (tokens) an AI model can process at once.
- Agentic workflow: The use of AI agents to perform a series of autonomous steps to reach a goal, rather than just generating a single response.
Original article
Google has released Gemini 4 Neon, a new member of the Gemini 4 family that brings frontier-level intelligence with fast end-to-end task completion for software engineering, knowledge work, and science.
Introducing Gemini 4 Neon! Pushing the frontier across domains and punching above its model class in a meaningful way. To me, Neon feels like the right balance of intelligence, speed, and usability. Available now across @GoogleAIStudio, the API, and more!
It is rolling out now in the Gemini API paid tier in Google AI Studio and Android Studio, in Gemini for business through the API and model dropdown, in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. Eligible members of the Fairwind Program will start getting a cyber defense version over the next few days.
Neon sits between Carbon, built for the hardest long-horizon projects, and Hydrogen, built for fast everyday execution. It costs $3 per million input tokens and $15 per million output tokens, with cached input at $0.15, and works with spreadsheets, PDFs, charts, audio and video in a 1 million-token context window.
Four thinking levels let teams trade intelligence for token spend, and at the default high level Google says it costs 32% less per task than Carbon. In Antigravity, Neon can act as an orchestrator directing parallel Hydrogen subagents, while in the Gemini app it now powers Tasks and deeper analysis.
Gemini 4 Neon is #1 on Cursorbench. Yet another Google Cursorbench mog.
Google reports a score of 61 on the Artificial Analysis Intelligence Index at xhigh, ranking third behind larger, slower models, along with first place on the Artificial Analysis Coding Agent Index at 58.6 and second on the Vals Index at 70.9%. In Google's comparison table, Neon scores 65.4% on CursorBench 4.1 against 58.1% for Claude Opus 5.5, plus 65.2% on Terminal-Bench 4.1 and 39.0% on SWE-sweep inside Antigravity.
Google DeepMind's Gemini 4 Neon debuts at #2 on the Vals Index (70.9%), 0.4 points behind Claude Fable 5.2 at less than half the cost per task. It also takes #1 on our Finance and Legal Indexes, Tax Agent and Legal Research.
Arena says Neon debuts second on its new Agent Arena Index with 1,336 points from about 4,200 real-world agent sessions, ahead of Carbon, Hydrogen and Argon, at a median cost of $2.82 per task. That is 53% less than Claude Opus 5.5, which Neon outranks. In the Text Arena, Google now holds the top three spots, with Neon third at 1,518.
Big news: Gemini 4 Neon (High) is the best @GoogleDeepmind model on the Agent Arena Index, ranking #2 overall and reshaping the Pareto frontier! With 1336 pts based on 4.2k real-world agentic sessions from Arena’s global community, it surpassed: #4 Carbon (High): 1303 pts.
The company says Neon keeps Carbon-level protections against severe misuse and prompt injection while declining far less legitimate work, and that it recorded its lowest fabrication rate yet in internal Antigravity testing. Google pitches it as frontier-level reasoning for complex, high-frequency workflows, delivered faster and at lower cost than its largest models.
Google continues to accelerate its release cadence with Gemini 4 Neon, delivering a small Intelligence gain over Gemini 4 Carbon at a ~17% lower Cost per Task @GoogleDeepMind has released Gemini 4 Neon, its fourth model in just seven weeks. Neon (xhigh) scores 61 on the…
I think this is a huge win for Google. A model listed below Carbon's standard price now outscores Claude Opus 5.5 on CursorBench, beats Carbon on Agent Arena, and lands near the top of the Artificial Analysis index.
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi