Google launches Gemini 4 Neon with Carbon quality at Flash speeds
Google launched Gemini 4 Neon for AI Pro subscribers, delivering Carbon-level coding performance at Flash speeds for $3 per million developer input tokens.
By Dillip Chowdary • Nov 18, 2026 • Source: 9to5Mac

At a glance
- Product
- Gemini 4 Neon
- Developer
- Target users
- AI Pro and AI Ultra subscribers
- Input price
- $3 per 1M tokens
- Output price
- $15 per 1M tokens
- Context window
- 1 million tokens
In this post
Google launched Gemini 4 Neon on November 18, 2026, delivering frontier reasoning performance comparable to or exceeding its flagship Carbon model while operating at Flash execution speeds. As detailed in 9to5Mac's report, the rollout expands Google's fourth-generation model tier by offering frontier-level reasoning for complex, high-frequency workflows faster and more cost efficiently. The new model permanently replaces Argon in the lineup, while legacy Gemini 3.1 Pro is officially discontinued for consumer and enterprise accounts.
This breakdown examines the architectural shifts, synthetic reasoning performance, benchmark scores, and production pricing across Google's restructured Gemini 4 portfolio. It is tailored for software engineers, enterprise architects, and paid subscribers evaluating whether to route high-volume automation workflows to Neon over heavier flagship options. Readers will find full technical metrics, benchmark rankings, token consumption rates, and practical implementation limits for production systems.
01Model picker restructuring and Argon replacement
Google reorganized its model selection matrix to balance latency, operational cost, and reasoning capability across consumer and enterprise accounts. Under the updated interface, Gemini 3.5 Flash-Lite powers the automated routing tier labeled Auto, while Hydrogen remains dedicated to everyday tasks. Neon replaces Argon to handle deeper technical analysis, positioning Carbon exclusively for the most demanding long-horizon computing jobs.
The restructuring permanently removes Gemini 3.1 Pro from the Gemini application, standardizing all advanced subscription workflows on fourth-generation weights. Paid users on AI Pro and AI Ultra plans gain immediate access to Neon for general reasoning, scripting, and document review. The consolidation establishes a four-tier operational ladder designed to prevent resource overprovisioning on standard developer workloads.
| Model | Target Tier | Role |
|---|---|---|
| Auto | Adaptive routing | Adapts to user needs using 3.5 Flash-Lite weights |
| Gemini 4 Hydrogen | Everyday help | Fast Flash replacement for everyday execution |
| Gemini 4 Neon | Deeper analysis | Complex reasoning and coding at Flash speeds |
| Gemini 4 Carbon | Code and complex work | Deepest reasoning for tasks testing model limits |
02Workspace document extraction and web research
Neon improves writing quality, knowledge extraction, and structural coherence when generating executive summaries and technical briefings from enterprise source materials. Google engineered the weights to read uploaded spreadsheets, PDF manuals, and images with heightened optical and structural precision. The model formats unstructured data into cleanly partitioned reports while adopting a natural, direct explanatory tone.
The model integrates directly with Google Workspace services to pull contextual information from personal and corporate accounts. Users can authorize Neon to reference messages within Gmail and files stored in Google Docs during research passes. This unified retrieval allows the model to surface real-time project metrics and calendar commitments without requiring manual export scripts.
For analytical tasks, Neon queries external web sources to maintain up-to-date competitive matrices and comparative pricing reports across vendors. The model targets multi-step administrative chores such as auditing utility spikes across recurring billing statements or compiling transaction receipts to secure vendor refunds. These background operations complete faster than on heavier frontier systems, conserving subscriber task allocations during high-frequency routines.
03Context capacity and four thinking levels
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
Neon features an expansive context window capable of ingesting up to one million tokens in a single execution context. This memory capacity enables engineering teams to feed entire software repositories, regulatory filings, or historical financial ledgers without aggressive chunking strategies. The extended window maintains consistent recall across document spans, reducing hallucination rates during comprehensive enterprise data extraction.
Like the lighter Hydrogen architecture, Neon exposes four distinct cognitive inference tiers designated as Low, Medium, High, and Extra High. Developers and application users can adjust these settings to balance internal computational deliberation against generation latency and token consumption budgets. The lowest tier delivers rapid responses for predictable queries, whereas Extra High unlocks extended reasoning paths for complex symbolic deductions.
The model occupies the midpoint between rapid surface execution and deep investigative engineering within the Gemini 4 family architecture. While Hydrogen handles everyday execution and Carbon remains dedicated to multi-hour frontier projects, Neon operates as a high-throughput production engine. Enterprise systems can deploy the intermediate weights across production codebases requiring consistent analytical depth without the overhead of Carbon.
04Benchmark evaluations across coding and reasoning
Benchmark evaluations position Neon near the top of competitive AI leaderboards across programming, mathematics, and agentic workflows. On the Artificial Analysis Intelligence Index at its Extra High thinking level, Neon achieved an overall score of 61, ranking third globally behind larger frontier models. The system also debuted in second place on the Agent Arena Index with an overall score of 1336 across more than 4,200 real-world sessions.
In specialized technical domains, Neon demonstrated dominant agentic task execution and structured numerical proficiency. It reached first place on both the Legal Research Bench and Tax Agent Bench, while securing the top rank on Surge AI across finance, accounting, and spreadsheet modeling tests. The model posted a 73.5 percent accuracy mark on Chartography, scored 75.8 percent on Agents' Last Exam v1.1, and led Zapier's AutomationBench at 56.3 percent.
On coding benchmarks, Neon captured first place overall on the Artificial Analysis Coding Agent Index with a score of 58.6. It recorded 65.2 percent on Terminal-Bench 4.1, 46.2 percent on FrontierCode 2.0 at high reasoning, and 70.5 percent on Terminal-Bench Science. Additionally, it achieved 84.0 percent on RiemannBench, 71.3 percent on HLE-Diamond, and earned $16,843 to win Andon Labs' Vending-Bench 2 business simulation.
05Token pricing structure and platform deployment
API pricing for Neon is set at $3 per million input tokens and $15 per million output tokens, making it twice as expensive as Hydrogen. However, promotional pricing for Argon and Carbon of $2 per million input and $10 per million output ends on November 30. Once Carbon and Argon rise to their standard rates of $4 input and $20 output, Neon will deliver a 25 percent cost reduction against those heavier configurations.
The model is deployed across Google consumer and developer environments, including the primary Gemini interface and Google Antigravity. Subscriber tiers on AI Pro and AI Ultra receive immediate access to test reasoning thresholds and automate complex code maintenance jobs. Technical organizations can leverage the lower pricing threshold to run continuous integration checks and enterprise support agents at scale.
Note: Access to Gemini 4 Neon in the Gemini web application and Google Antigravity developer workspace is restricted to active Google AI Pro and AI Ultra subscribers. Legacy Gemini 3.1 Pro endpoints are permanently deprecated as part of this rollout, requiring existing integrations to migrate to the updated Gemini 4 model family.
Frequently asked questions
Which subscription plans include access to Gemini 4 Neon?
Gemini 4 Neon is available directly in the Gemini application and Google Antigravity for paid Google AI Pro and AI Ultra subscribers.
How does Gemini 4 Neon compare to Carbon and Flash models?
Neon delivers reasoning and coding performance comparable to or exceeding Carbon while executing at Flash speeds with a one million-token context window.
What are the API pricing rates for Gemini 4 Neon?
Neon costs $3 per million input tokens and $15 per million output tokens, which will be 25 percent cheaper than standard post-promotional Carbon pricing.
Which previous models does Gemini 4 Neon replace?
Neon replaces Argon in Google's model picker, while legacy Gemini 3.1 Pro has been permanently retired.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement