Gemini 4 Neon ships with two-model relay, tops engineering benchmark
What happened
On November 19, Google released Gemini 4 Neon for complex coding, knowledge work and science tasks, rolling it out gradually to developers, enterprises and paid users. At the xhigh level it scored 61, third on the Artificial Analysis overall leaderboard, and at the default high setting its cost per task is 32% lower than Carbon's. API pricing is $3 per million input tokens and $15 per million output tokens. llm-gemini 0.38 shipped the same day. Reports say Antigravity added an Orchestrator, with Neon planning and Hydrogen executing, handing off within a single session and opening to Pro subscribers. Earlier reports said a fourth Gemini 4 model would arrive within four weeks, later ones within seven. Earlier reports put generation speed at 127 versus 46 tokens per second, later ones at nearly double, with no explanation of the gap. On November 20, reports showed Neon topping the Models Engineering Index engineering task board at 69.5, ahead of Fable 5.2 at 65.8 and Opus 5.5 at 63.1.
Written by AI from the coverage · updated 15 hours ago
Developments
- Nov 20 07:35 · 1 reportGemini 4 Neon is the new #1 on the Models Engineering IndexHacker News front page
- Nov 19 08:29 · 1 reportGemini 4 Neon looks like an Opus 5.5 competitorHacker News front page
- Nov 19 07:59 · 1 reportllm-gemini 0.38Simon Willison
- Nov 19 05:00 · 5 reportsGoogle releases Gemini 4 Neon for complex coding, knowledge work and science tasksGoogle Blog · Gemini
Coverage
Follow the reports to see the story from different sides.
- Computing Life · Share · YagePickGemini 4 Neon: how Google hands off between two models in one session to cut costs
Google added an Orchestrator (Neon) to Antigravity. Gemini 4 Neon plans and Hydrogen executes, with the handoff happening inside a single session and opened to Pro subscribers.
- Hacker News front pageGemini 4 Neon is the new #1 on the Models Engineering Index
Gemini 4 Neon 在 Models Engineering Index(Mei)工程任务评测中以 69.5 分排名第一,领先 Anthropic 的 Fable 5.2(65.8 分)和 Opus 5.5(63.1 分)。该榜单按模型在工程任务上的实际表现排序,同榜还有 OpenAI 6.2 Astra(60.5 分)与 Google 4 Hydrogen(57.2 分)。
- The DecoderPickGoogle releases Gemini 4 Neon, beating Carbon on most benchmarks with nearly double the speed and lower standard pricing
Google DeepMind released Gemini 4 Neon, its fourth model in seven weeks. It beats Carbon on most benchmarks and generates nearly twice as fast. The model takes a 1M-token context window across text, image, audio and video, and ranks third on the Artificial Analysis Intelligence Index with a score of 61.
- Hacker News front pageGemini 4 Neon looks like an Opus 5.5 competitor
Agent Duel 的建模分析预测,若 Gemini 4 Neon 的能力与 Gemini 4 Carbon 相同,其榜单表现将接近 Opus 5.5。Neon 今天发布,基准成绩小幅提升,在 OpenRouter 上的生成速度达到 127 tokens per second,Carbon 为 46 tokens per second。
- Simon Willisonllm-gemini 0.38
Simon Willison 发布 llm-gemini 0.38,并比较今天 Google 推出的 Gemini 4 Neon 与系列其他模型的价格和生成表现。
- Google DeepMindPickGoogle DeepMind releases Gemini 4 Neon for complex coding, knowledge work and science tasks
Google DeepMind released Gemini 4 Neon today, aimed at complex software engineering, knowledge work and scientific workflows. It is rolling out to developers, enterprises and paid subscribers.
- The DecoderPickGoogle releases Gemini 4 Neon, its fourth Gemini 4 model in four weeks
Google released Gemini 4 Neon, available today through the API at $3 per million input tokens and $15 per million output tokens. The model has a context window of about one million tokens and can output up to one million tokens. It accepts text, image, video, audio and PDF input but does not generate images or audio.
- Google Blog · GeminiPickGoogle releases Gemini 4 Neon for complex coding, knowledge work and science tasks
Google released Gemini 4 Neon, aimed at complex software engineering, knowledge work and scientific workflows, rolling out today to developers, enterprises and paid subscribers. At the xhigh thinking level it scores 61 on the Artificial Analysis Intelligence Index, third overall; at the default high level, cost per task is 32% lower than Carbon.
Heat over time
Heat now 24·Comparable peak 35(Nov 20 09:00)·Comparable change over 24 hours -16%
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.