Gemini 4 Neon release
Google introduced Gemini 4 Neon, which it says delivers frontier-level intelligence with fast task completion for software engineering, knowledge work and science, and began rolling it out the same day.1234 Per Latent Space, Neon is the fourth Gemini 4 model in about seven weeks, and the earlier Argon model is now retired.56 Neon is priced at $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 and a 1 million-token context window.78
According to Google, its Neon model takes third place overall on the Artificial Analysis Intelligence Index, scoring 61 when set to xhigh thinking, and comes second on the Vals Index.9101112 Google also claims a 32% lower cost per task than Carbon at its default high setting, and Latent Space reports that Neon matched Carbon's 60 on the Intelligence Index at high effort for $3.74 per task.131415 Independent results are mixed: Latent Space says Neon debuted second on Agent Arena with 1,336 points while Carbon still leads the Text Arena, and on Vals Code Migration Neon scored below both Carbon and Hydrogen.161718
Simon Willison says the Gemini 4 pricing was already confusing and Neon made it worse, because Carbon is currently half price, making Neon the most expensive of the three until Carbon's price doubles on December 1st.192021 Willison also says his pelican test images for Neon look better than Hydrogen's and worse than Carbon's.22
A micjm.com analysis projects that Neon, assuming Carbon-level performance at the 127 tokens per second it saw via OpenRouter, would reach an Elo of 1,066 on its Agent Duel leaderboard, just below Opus 5.5 at 1,078, though the 90% interval for the difference spans -94 to +68.2324
Sources quoted
Today, we’re introducing Gemini 4 Neon — bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.
— Google DeepMind, Introducing Gemini 4 NeonGemini 4 Neon is rolling out today in the Gemini API in Google AI Studio
— Google DeepMind, Introducing Gemini 4 NeonToday, we’re introducing Gemini 4 Neon — bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.
— Google AI, Gemini 4 NeonGemini 4 Neon is rolling out today in the Gemini API in Google AI Studio
— Google AI, Gemini 4 NeonGoogle DeepMind shipped Gemini 4 Neon, its fourth Gemini 4 model in about seven weeks.
— Latent Space, [AINews] Gemini 4 Neon: GDM's Opus/Sol Contenderthe 1.5 month old Argon is now retired
— Latent Space, [AINews] Gemini 4 Neon: GDM's Opus/Sol ContenderThe model is priced at $3 per million input tokens ($0.15 per million cached input tokens, a 95% discount) and $15 per million output tokens, with a 1 million-token context window.
— Google DeepMind, Introducing Gemini 4 NeonThe model is priced at $3 per million input tokens ($0.15 per million cached input tokens, a 95% discount) and $15 per million output tokens, with a 1 million-token context window.
— Google AI, Gemini 4 NeonNeon ranks ahead of comparable models and #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level
— Google DeepMind, Introducing Gemini 4 NeonOn the Vals Index, Neon ranks #2 and lands on the Pareto frontier.
— Google DeepMind, Introducing Gemini 4 NeonNeon ranks ahead of comparable models and #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level
— Google AI, Gemini 4 NeonOn the Vals Index, Neon ranks #2 and lands on the Pareto frontier.
— Google AI, Gemini 4 Neonit pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon.
— Google DeepMind, Introducing Gemini 4 NeonAt high effort, Neon matched Carbon’s 60 on the Intelligence Index for $3.74 per task.
— Latent Space, [AINews] Gemini 4 Neon: GDM's Opus/Sol Contenderit pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon.
— Google AI, Gemini 4 NeonNeon debuted second with 1,336 points across about 4,200 sessions.
— Latent Space, [AINews] Gemini 4 Neon: GDM's Opus/Sol ContenderCarbon still leads the separate Text Arena.
— Latent Space, [AINews] Gemini 4 Neon: GDM's Opus/Sol ContenderOn Code Migration it averaged 160 turns, fewer than Carbon’s 234 or Hydrogen’s 574, and scored below both.
— Latent Space, [AINews] Gemini 4 Neon: GDM's Opus/Sol ContenderI wrote recently about how confusing the pricing for the Gemini 4 series was - Neon has somehow made this even worse.
— Simon Willison, llm-gemini 0.38right now Neon ($3/million input, $15/million output) is the most expensive of the three
— Simon Willison, llm-gemini 0.38I guess this will make more sense when Carbon doubles to $4/$20 on December 1st.
— Simon Willison, llm-gemini 0.38the Neon pelicans do look better than Hydrogen and worse than Carbon
— Simon Willison, llm-gemini 0.38we would project an Elo on the Agent Duel leaderboard of 1,066, just below Opus 5.5 at 1,078 in our later analysis snapshot.
— micjm.com, Gemini 4 Neon looks like an Opus 5.5 competitorThe 90% interval for the projected Elo difference between Gemini 4 Neon and Opus 5.5 is -94 to +68
— micjm.com, Gemini 4 Neon looks like an Opus 5.5 competitor
Who reported it
- Simon Willisonown facts: Neon is priced above Carbon and Hydrogen for now. The author calls the Gemini 4 pricing confusing and says the announcement ignores Argon.
- Latent Spacerepeats Google: Midtier Gemini 4 model at $3/$15 per million tokens, 1M context, four effort levels; rolling out to API, apps, and day-one integrations like Cursor and GitHub Copilot.
- Hacker Newsannouncement: Gemini 4 Neon
- Google DeepMindannouncement: Neon launches today in Gemini API, AI Studio, Gemini for business, Antigravity, and the Gemini app, claiming top benchmark ranks at lower cost per task.
- micjm.comown facts: Released today with incremental benchmark gains and much faster serving speed, reportedly 127 tokens per second on OpenRouter.
- Google AIannouncement: Rolls out today across Gemini API, AI Studio, Antigravity, Gemini for business, and Gemini app; claims top benchmark rankings at lower cost.
- Hacker Newsown facts: Gemini 4 Neon is the new #1 on the Models Engineering Index
- Hacker Newsown facts: Gemini 4 Neon looks like an Opus 5.5 competitor
- AI Explainedcommentary: Cited as Gemini catching up quickly, alongside 4 Carbon; the video links Google's Gemini 4 Neon announcement.
Facts and takes
Facts: 52 checked facts from 8 sources
- Google released Gemini 4 Neon
- Gemini 4 Neon: $3 per 1M input tokens
- Gemini 4 Neon: $15 per 1M output tokens
- Gemini 4 Neon: $0.15 per 1M cached input tokens
- Gemini 4 Neon: 70.9% on Vals Index
- Gemini 4 Neon: 1M context window
- Gemini 4 Neon is available on Gemini Live
- Gemini 4 Neon is available on Gemini API
- Gemini 4 Neon: 61 on Artificial Analysis Intelligence Index (xhigh thinking level)
- Gemini 4 Neon: 1336 on Agent Arena Index
- Gemini 4 Neon: 73.5% on Chartography
- Gemini 4 Neon: 75.8% on Agents’ Last Exam v1.1
- Gemini 4 Neon: 56.3% on AutomationBench
- Gemini 4 Neon: $16,843 on Vending-Bench 2
- Gemini 4 Neon: 70.5% on ProgramBench
- Gemini 4 Neon: 84.0% on RiemannBench
- Gemini 4 Neon: 71.3% on HLE-Diamond
- Gemini 4 Neon: 58.6 on Artificial Analysis Coding Agent Index
- Gemini 4 Neon: 65.2% on Terminal-Bench 4.1
- Gemini 4 Neon: 46.2% on FrontierCode 2.0 (default high setting)
- Gemini 4 Neon: 12.8% on SWE-sweep (mini-SWE harness)
- Gemini 4 Neon: 39.0% on SWE-sweep (Google Antigravity harness)
- Gemini 4 Neon: 76% on AutoBaxBench (hard tier)
- Gemini 4 Neon: 51.26% on WebExploitBench v1.2
- Gemini 4 Neon: 90.5% on SEC-bench Pro
- Gemini 4 Neon: 70.7% on CWE-bench v2
- Gemini 4 Neon is available on Google AI Studio
- Gemini 4 Neon is available on Gemini for business
- Gemini 4 Neon is available on Google Antigravity
- Gemini 4 Neon is available on Fairwind Program
- Gemini 4 Neon: cost per task down 32% vs Gemini 4 Carbon
- Gemini 4 Neon: cached input price down 95%
- Gemini 4 Neon: thinking levels 4
- Gemini 4 Neon: handles context compaction reliably in long sessions
- Gemini 4 Neon: directs parallel subagents as an Orchestrator in Google Antigravity
- Gemini 4 Neon: $3 input (paired with $15 output)
- Gemini 4 Neon: $15 output (paired with $3 input)
- Gemini 4 Neon: 61 on Artificial Analysis Intelligence Index
- Gemini 4 Neon: 61 on Artificial Analysis Intelligence Index (xhigh)
- Gemini 4 Neon: 63 on 105 replanted bugs (xhigh)
- Gemini 4 Neon: 60 on Artificial Analysis Intelligence Index (high effort)
- Gemini 4 Neon: 33.4% on GDP.pdf (high)
- Gemini 4 Neon: 28.0% on GDP.pdf (xhigh)
- Gemini 4 Neon: 1,336 on Agent Arena
- Gemini 4 Neon: 63.7% on CursorBench 4.1 (default effort)
- Gemini 4 Neon: 68.1 on Models Engineering Index
- Gemini 4 Neon is available on Antigravity
- Gemini 4 Neon is available on OpenRouter
- Gemini 4 Neon: maximum output tokens 1M tokens
- Gemini 4 Neon: throughput 127 tokens per second
- Gemini 4 Neon: four effort levels
- Gemini 4 Neon: three reasoning levels: low, medium, high
Takes: 5 views from 3 holders · 3 positive · showing 3
- +Latent Space:
Because of how distillation, RL, and benchmarking works, Neon technically outperforms Carbon by some measures despite being a smaller model
- ·micjm.com: Quote not reprinted: this page already repeats as much of this article as we allow.
- +Jeremy Howard :
Jeremy Howard praised Neon while saying he was reverting Solveit’s Astra and Sol options to their pre-6.2 versions.
Timeline
- · Hacker News: Gemini 4 Neon
- · Google AI: Gemini 4 Neon
- · Google DeepMind: Introducing Gemini 4 Neon
- · Simon Willison: llm-gemini 0.38
- · micjm.com: Gemini 4 Neon looks like an Opus 5.5 competitor
- · Hacker News: Gemini 4 Neon looks like an Opus 5.5 competitor
- · Latent Space: [AINews] Gemini 4 Neon: GDM's Opus/Sol Contender
- · AI Explained: Anthropic IPOs, GPT 6.2 Astra Arrives, But There’s a Bigger Story
- · Hacker News: Gemini 4 Neon is the new #1 on the Models Engineering Index