Model

Google opens Gemini 4 Neon to all developers at $3 per million input tokens

Google's fourth Gemini 4 model in five weeks pairs top-three benchmark scores with mid-tier pricing, resetting the margin math for AI software vendors in the very week Anthropic's record IPO slid back toward its $83 offer price.

Richard TangRichard Tangthree-minute read
Share
Sundar Pichai, Google's chief executive, speaks into a microphone on stage at an event in Hanoi
Sundar Pichai, Google's chief executive, at a developer event in Hanoi in December 2015.

A $3 model, and a weekend on Antigravity

Google moved Gemini 4 Neon, the fourth model in its Gemini 4 series since Argon opened it on 13 October, to general availability on 18 November: any developer with a paid Gemini API key can call the stable endpoint at $3 per million input tokens and $15 per million output, with cached input at $0.15, a 95% discount 26.

The weekend turn came on Antigravity, Google's−1.04% — Google, down 1.04 percent today agentic development platform 13. The story's most-viewed post announced that every user's Gemini quota had been reset and claimed caching improvements now buy 22% more work within quota 1. Google serves Neon there to subscribers of Google AI Pro and Ultra, its paid consumer tiers, alongside AI Studio, the Gemini app and business plans 3.

What $3 per million tokens buys

Neon takes text, images, audio, video and PDFs into a single one-million-token window shared by input, thinking and output, with four reasoning levels from low to xhigh, high by default 23. The input rate is half again the $2 that OpenAI's$1.18T — OpenAI, private, latest valuation $1.18T GPT-6.2 Sol and Anthropic's$350B — Anthropic, private, latest valuation $350B Claude Sonnet 5.5 charge, and a quarter under Claude Opus 5.5's $4 4.

Neon's $3 input rate undercuts the flagship models it matches on benchmarks

A top-three model priced below both flagships' standard rates

$0/1M input$2/1M input$4/1M input$6/1M input$8/1M inputGemini 4 HydrogenGPT-6.2 SolClaude Sonnet 5.5Gemini 4 NeonClaude Opus 5.5GPT-6.2 Astra$8/1M input$3/1M input
Data
Input price
Gemini 4 Hydrogen$1.5/1M input
GPT-6.2 Sol$2/1M input
Claude Sonnet 5.5$2/1M input
Gemini 4 Neon$3/1M input
Claude Opus 5.5$4/1M input
GPT-6.2 Astra$8/1M input
Standard short-context API input rates, US dollars per million input tokens, as of 21 November 2026.2,4

The independent check: third overall, and faster

Artificial Analysis, the third-party benchmark tracker, scored Neon 61 at its xhigh thinking level: third overall, behind GPT-6.2 Astra and Claude Fable 5.2, and a point above Google's own Carbon at high 5. That run cost $6.44 per Intelligence Index task, the tracker's average cost for one pass through its benchmark suite, and took about 21 minutes 5.

Thai Tran, AI product lead at Glean$6.46B — Glean, private, latest valuation $6.46B and an early customer quoted in Google's announcement, reports Carbon-quality work at three times the speed and 60% of the cost 4. The cheaper bill comes from the rate card, not a leaner model: Neon used about 15% more output tokens per task than Carbon 5.

The software around the model is now the bottleneck

On Google's SWE-sweep coding test, the model card reports Neon at 39.0% inside the provider harness, Antigravity's agent software, and 12.8% under the bare mini-SWE setup; Claude Opus 5.5 managed 10.1% inside the same Antigravity harness and 3.6% under mini-SWE 3. The software around the model tripled the score 34. A leaderboard rank prices the model; the agent software decides how much of it becomes finished work. The caching work behind the weekend's 22% claim is that kind of engineering 1, and it is where the margin on a $3 model is now earned.

Antigravity's agent harness tripled both models' coding scores

The software around the model moved the score more than the model did

  • Inside Antigravity's harness
  • Bare mini-SWE setup
0%10%20%30%40%Gemini 4 NeonClaude Opus 5.510.1%39%3.6%
Data
Inside Antigravity's harnessBare mini-SWE setup
Gemini 4 Neon39%12.8%
Claude Opus 5.510.1%3.6%
SWE-sweep scores reported on Google's Gemini 4 Neon model card, percent of tasks solved, run inside Antigravity's agent harness and under a bare mini-SWE setup.3,4

Anthropic gave back its debut gain in two sessions

Anthropic priced at $83 on Wednesday, raised a record $81 billion, opened Thursday's debut at $111.50 and closed at $97.17, then dropped more than 12% on Friday to about $85 by midday while Alphabet, alone among the megacap technology names, gained 1.2% 7.

Two sessions took Anthropic from $111.50 back to about $85

A 17% debut gain nearly erased in one session

$80$90$100$110$120IPO price, 18 NovOpen, 19 NovClose, 19 NovMidday, 20 Nov$85Almost back to the $83 offer price
Data
Anthropic
IPO price, 18 Nov$83
Open, 19 Nov$111.5
Close, 19 Nov$97.17
Midday, 20 Nov$85
Anthropic share price in US dollars: IPO pricing on 18 November, opening and closing prints of the 19 November debut, and midday on 20 November.7

The posts frame the stakes bluntly: "It might be over for OpenAI" 9; "the race is still wide open" 10. A leak already shows gemini-cobalt-thinking-preview-001 in AI Studio, which Google has not confirmed, keeping the elemental cadence going 8.

Frontier coding at $3 a million tokens 2 moves the constraint from the model to the software around it, and the invitation for builders is to do that work. The next dates to watch are Google's own: whether it confirms the Cobalt sighting 8, and what the price ladder looks like after 30 November, when Carbon's $2 introductory rate ends and its input price doubles to $4 2.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading GOOGL

Reader comments

0 comments

    Model

    House report says the model behind SpaceX's $60 billion Cursor deal was built on China's Kimi K2.5

    SpaceX paid about $60 billion in stock for Cursor in August, and a House report now says its flagship model rests on Chinese open weights. The same day, rival Cognition entered talks at $75 billion, two months after Cursor's own backers led its $48 billion round.

    Model

    Mira Murati's Thinking Machines ships a 26-billion-parameter model that runs offline on a MacBook

    The 26-billion-parameter open-weights agent shipped hours after OpenAI said attackers had extracted GPT-6 Astra's hidden reasoning at a 54% success rate and asked Congress for new enforcement powers. Together the two moves frame the scarcity bet behind frontier token prices and the market's newest trillion-dollar listing.

    Model

    OpenAI says attackers copied GPT-6 Astra's hidden reasoning on 54% of tries; its fix cuts that to 0.2%

    OpenAI's November 20 disclosure puts a hard number on how copyable a frontier model was, and pairs the fix with a call for federal enforcement. The labs it names, Moonshot AI and DeepSeek among them, are where the money moves.

    Model

    Amazon's $188 Billion Anthropic Stake Swings $1.9 Billion for Every $1 of Anthropic Stock

    Two trading days turned Amazon's largest AI holding into a daily mark-to-market item, and Anthropic's second-session slide shows how fast the gain Amazon already booked can reverse.

    Model

    HeyGen sells Avatar 6 while Tavus's Griffin scores 3.83 against a human's 3.92 and still has no price

    HeyGen put Avatar 6 on general availability on November 17 with a public API and first place in New Measure's blind Avatar Arena; Tavus's Griffin, which Tavus says is the highest score ever recorded for an AI on NVIDIA's face-to-face benchmark, is still a preview with no price. Revenue accrues to HeyGen today because Avatar 6 is the model on sale; Tavus's best model is not.

    Model

    Pika's Enact lets creators direct AI video mid-scene, with answers in under a second

    Enact, posted November 18, lets a creator drag an AI actor mid-scene, talk to the cast by name and redo the camera after locking a performance. Pika open-sourced the model underneath and kept the directing layer closed, a bet that value in AI video is moving from model quality to workflow.

    Subscribe to TheInference for free