Gemini 4 Neon
Confirmed
Importance: major (4 of 5)The takeaway
On 18 November 2026 Google released Gemini 4 Neon, its fourth Gemini 4 model in seven weeks, positioned between Carbon and Hydrogen for fast end-to-end coding, knowledge-work and science tasks.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Major (4 of 5)
- Last verified
- 19 November 2026
Your AI and this story
- GPT-6.2 Astra171 days after its cutoff
- Claude Opus 5.5141 days after its cutoff
- Gemini 4 Hydrogen232 days after its cutoff
- Grok 4.7171 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 141 days before it.
Key facts
- Availability (Nov 18): Gemini API id gemini-4-neon (paid tier), Gemini app for AI Pro and Ultra (replaces Argon in the app), Antigravity, AI Studio, Gemini for business, Android Studio; GitHub Copilot added it the same day; Fairwind cyber-defender access over the following days
- Price: $3 input / $15 output per 1M tokens, cached input $0.15; 1M context; four thinking levels (default high, top xhigh)
- Artificial Analysis (independent): Intelligence Index 61 at xhigh, #3 behind GPT-6.2 Astra (max) and Claude Fable 5.2 (max), Carbon (high) 60; $6.44 per index task vs $7.76 for Carbon; ~110 tok/s, ~21 min per task; Coding Agent Index 59 in Antigravity CLI (#1) vs 58 for Opus 5.5 (max) and Fable 5.2 (xhigh); AA-Briefcase 1700 Elo
- Arena (independent): #2 on the Agent Arena Index at 1336 (4.2k real sessions), ahead of #3 Claude Opus 5.5 (Max) at 53% lower median cost per task ($2.82); Fable 5.2 (Max) #1
- Vendor numbers: Vals Index 70.9% (#2); Terminal-Bench 4.1 65.2%; FrontierCode 2.0 46.2%; SWE-sweep 39.0% in Antigravity vs 12.8% in the mini-SWE harness (the harness tripled the score, The New Stack); Terminal-Bench Science 70.5%; RiemannBench 84.0%; HLE-Diamond 71.3%; Zapier AutomationBench 56.3% (#1); Vending-Bench 2 $16,843
- Cyber (vendor): AutoBaxBench hard 76%, WebExploitBench v1.2 51.26%, SEC-bench Pro 90.5%, CWE-bench v2 70.7%
- Safety (vendor): Carbon-level misuse protections; 0.09% misaligned-action rate on Google’s agentic safety suite; 99.2% on agent honesty; 2.45% shortcuts on a coding-cheating eval (lowest Gemini 4); 0.0% attack success on Gray Swan IPI at highest thinking; model card published
- Cadence (Arena): Argon (Sept 30) → Carbon 28 days → Hydrogen 15 days → Neon 6 days; Demis Hassabis’ launch post 734k views, Google DeepMind’s 707k; HN 865 points
What happened
Google DeepMind shipped Neon six days after Hydrogen. It is pitched as frontier quality at mid-tier price and speed, and replaces Argon as the model AI Pro subscribers see in the Gemini app. Independent boards placed it just below the newest OpenAI and Anthropic flagships overall and level with or ahead of them on agentic coding cost-effectiveness.
Why it matters
Release gaps at Google have shrunk to days, and the price of near-frontier capability keeps dropping: Neon matches Carbon at about half the cost per task at high effort (Artificial Analysis). Several of Google’s best scores depend on its own Antigravity harness, a reminder that agent results now measure model plus harness.
Sources
6 sources from 6 sites. Numbers match the chips in the text.
6 sources: 2 primary, 2 press, 2 reactions
Primary
- Google: Introducing Gemini 4 Neon (Koray Kavukcuoglu, Nov 18)blog.google, official
- Google DeepMind: Introducing Gemini 4 Neondeepmind.google, official
Press
- Artificial Analysis: Gemini 4 Neon intelligence gainartificialanalysis.ai, press
- 9to5Google: Gemini 4 Neon costs $3/$15, replaces Argon in the app9to5google.com, press
Reactions
- Arena on X: Neon #2 on Agent Arena Indexx.com, discussion
- Hacker News discussion (865 points)news.ycombinator.com, discussion
Changes
- Filed