Gemini Neon
by Google · Model · Language & code models
promising
Close to Fable on agent rankings at half the price, but only a day old.
Gemini 4 Neon shipped, the fourth Gemini 4 model in about seven weeks; Argon is retired
AccessAPIPrice$3 in · $15 out per M tokensContext1,049K tokens
Why we say so
Vals puts Neon second at 70.9%, about 0.4 points behind Fable 5.2 at under half the cost per task, and Arena debuted it second on agents. Cursor measured 63.7% on its benchmark, up from Hydrogen, and Jeremy Howard praised it. Evidence is early: Carbon still leads Arena's text ranking, Simon Willison's pelican SVGs were worse than Carbon's, and Fahd Mirza found a demo with patchy features. Its strongest benchmark wins over Claude and GPT are Google's own table.
Ranks #2 for agents · #2 for chat · #3 for coding
Evidence
- Nov 19Commenter made progress on an ML side project with Neon after Opus and Fable stalled; single anecdote
- Nov 19In Hermes Agent, one mega prompt with Neon managing helper agents finished in about 36 minutes; a checker passed six of six headless-browser checks
- Nov 19At high effort Neon matched Carbon's 60 on the Intelligence Index at $3.74 per task
- Nov 19Ranks #1 on Artificial Analysis Coding Agent Index (58.6); #1 on Legal Research Bench, Tax Agent Bench, AutomationBench, Vending-Bench 2
- Nov 19Performs just under Fable overall on Vals comparison, slightly ahead of Opus 5.5 and above Astra, Sol 6.2 and Carbon
Show all 16
- Nov 19Slightly ahead of Opus 5.5 on Vals comparison; commenter says it feels as good as Opus 5.5
- Nov 19Better and faster than GPT-6.2 Sol on Vals but about 2x more expensive; best latency for its quality level
- Nov 19Announced hours earlier as the middle Gemini 4 model, built for deep reasoning in software engineering, knowledge work and science
- Nov 19Scored 63.7% on CursorBench 4.1 at default effort, up from Hydrogen's 57.3%
- Nov 19On Vals Code Migration, Neon averaged 160 turns versus Carbon's 234 and Hydrogen's 574, but scored below both
- Nov 19Commenter finds it as good as Gemini Argon but faster and (they think) cheaper
- Nov 19Google's table has Neon leading most rows (agentic coding, knowledge work, science, math, long context, cybersecurity) against Claude 5 and GPT 6.2
- Nov 18Gemini 4 Neon released, a mid-tier model between Carbon and Hydrogen for complex coding, reasoning and enterprise workflows
- Nov 11#2 on Arena text (1556)
- Nov 11#1 on Arena text (1534)
- Nov 11#2 on Arena text (1530)