
Tech • AI • Robotics • Game
⏱ ≈ 3 min read · 10-min video — 7 min saved
Google’s Gemini 4 Neon is rolling out across Gemini and AI Studio as a faster, lower-cost reasoning model that in early hands-on testing often matched or outperformed Gemini 4 Carbon on coding, game creation and animation tasks.
Google describes Gemini 4 Neon as a fast reasoning model that delivers quality close to Gemini 4 Carbon for coding, reasoning and production-agent work, but at higher speed and lower cost. It is positioned between Gemini 4 Hydrogen, aimed at high-volume everyday use, and Carbon, which is intended for the hardest long-horizon problems.
Neon is appearing inside paid Gemini accounts priced at $20 per month, as well as in Google AI Studio. It has also surfaced in other Google tooling, indicating a broader rollout rather than a limited preview.
At the time of testing, no official benchmark sheet had been published for Gemini 4 Neon. That leaves early comparisons largely dependent on direct use cases rather than standardized leaderboard scores.
In a side-by-side visual generation test, Carbon produced an image with more perceived depth and texture, especially in clothing, while Neon delivered a more realistic face and cleaner line work. The outcome was notable because the preferred result on several technical details came from Neon, despite Carbon being positioned as the stronger model.
Neon was used to build a simple playable 3D foraging game set in a forest glade. The first version had pathing and collision issues, including a character failing to walk correctly on a bridge and clipping through the map while foraging, but follow-up fixes were handled quickly, leaving a small but visually appealing game.
In a recurring test using the same prompt for both models, each was asked to create a top-down 3D paintball capture-the-flag game using Godot and Blender. The Neon version was judged sharper and more visually polished, with stronger camera movement, while the Carbon version felt slightly easier to play. On looks alone, Neon came out ahead.
A separate Star Fox-style space shooter test showed a narrower but clearer win for Carbon. Both models produced functional games, but Carbon delivered somewhat better environmental polish, especially in water rendering, suggesting it still holds an advantage on some higher-fidelity outputs.
In a 90-second animation prompt involving a couple chased by a wolf through doors that changed visual style, both models followed instructions and shifted styles as requested. However, Carbon struggled heavily with limbs and awkward body motion, while Neon showed better anatomy and more ambitious framing, including dynamic wolf shots. Neither result was flawless, but Neon was the stronger performer.
In hands-on testing, Neon was estimated to be about 2.5 times faster than Carbon. That matters because even in cases where Carbon produced slightly better results, Neon still delivered usable output much more quickly, improving iteration speed for practical workflows.
Hydrogen appears best suited as the default everyday model. Neon looks positioned to handle most higher-capability work for typical users, especially where speed matters. Carbon still has a role for situations demanding the very best output on difficult, long-horizon tasks, but Neon’s strong performance makes that upgrade decision less obvious.
Gemini 4 Neon appears to be more than a mid-tier compromise: it is emerging as a strong general-purpose model that can challenge Carbon on quality while decisively beating it on speed. If broader testing confirms these early results, Neon could become the most practical choice in Google’s current model lineup.
Ask a question