Google launched Gemini 4 Neon, its fourth AI model in seven weeks, across the Gemini API, AI Studio, Vertex AI, the Gemini app and Antigravity for paid Pro and Ultra subscribers. Positioned between Gemini 4 Hydrogen and Carbon, the model features four reasoning effort levels from low to xhigh, accepts text, image, video and audio inputs with a 1M token context window, and outputs up to 1M tokens at roughly 110 tokens per second, up from 57 for Carbon.
Independent evaluations by Artificial Analysis and Vals AI show Neon matching or beating larger frontier models at roughly half the cost per task. In Antigravity CLI, Neon at xhigh effort tops the Artificial Analysis Coding Agent Index with a score of 59 at an average cost of $6.86 per task, beating Anthropic’s Claude Fable 5.2 and Claude Opus 5.5, which both scored 58 at $15.47 and $12.61 per task. On the Vals Index, Neon debuted at No. 2 with 70.9%, trailing Claude Fable 5.2 by 0.4 percentage points at less than half the task cost while taking first place on the Finance, Legal, Tax Agent and Legal Research indexes.
On the SWE-sweep coding benchmark, Neon’s score rises from 12.8% using a standard mini-swe-agent harness to 39.0% when run inside Google’s Antigravity environment, where it can delegate tasks to smaller Flash and Hydrogen sub-agents. Priced at $3 per 1M input tokens, $15 per 1M output tokens and $0.15 for cached input, Neon ranks third overall with a score of 61 on the Artificial Analysis Intelligence Index, behind OpenAI’s GPT-6.2 Astra and Claude Fable 5.2.
