Back to News
ProductAI Understanding briefing

Google launches Gemini 4 Neon for complex coding at lower cost

Google introduced Gemini 4 Neon, a new AI model aimed at complex coding and knowledge‑work tasks, priced at $3 per million input tokens and $15 per million output tokens, with a 1 million‑token context window.

4 min readRead the linked source
Source-provided image accompanying Google launches Gemini 4 Neon for complex coding at lower cost
Source referenceSource recorded
Publisher
tbreak.com
Source type
Linked source — primary-source status has not been established.
ContextUnderstand this in 60 seconds

Key terms

API (Application Programming Interface)
A structured way for one software system to send requests to and receive responses from another system.
Prompt Engineering
Designing prompts to improve output quality, reliability, and controllability.
Context Window
The maximum amount of input tokens a language model can process at once.

What happened

Google announced Gemini 4 Neon, a new tier in its Gemini 4 family positioned between the high‑performance Carbon model and the fast‑response Hydrogen model. Neon is marketed for complex coding and knowledge‑work tasks. Through the Gemini API and Google AI Studio, Neon is available to Google AI Pro and Ultra subscribers, as well as via the Gemini app and Google Antigravity for paid users. Pricing is set at $3 per million input tokens and $15 per million output tokens, with cached inputs discounted to $0.15 per million tokens. The model supports up to 1 million tokens of context. Benchmarks released by Google place Neon first on the Artificial Analysis Coding Agent Index (score 58.6) and second on several other internal rankings, while claiming a 32% lower cost per task than Carbon at comparable settings.

Google introduced Gemini 4 Neon on 18 November as part of the Gemini API and Google AI Studio. The model is positioned between the existing Carbon (high‑capacity, high‑cost) and Hydrogen (fast, low‑cost) tiers.

Neon targets complex coding and knowledge‑work tasks, offering a 1 million‑token —significantly larger than most current models. Pricing is $3 per million input tokens and $15 per million output tokens, with cached inputs at $0.15 per million tokens, a 95% discount on the standard input rate.

Access is provided through the Gemini API, the Gemini app, and Google Antigravity for paid Google AI Pro and Ultra subscribers. The rollout also includes a limited early‑access programme (Fairwind) for defenders, mirroring the earlier Gemini 4 Argon rollout.

Google’s internal benchmarks place Neon first on the Artificial Analysis Coding Agent Index (score 58.6) and second on the Vals Index (70.9%). It scored 65.2% on Terminal‑Bench 4.1 and 46.2% on FrontierCode 2.0 at its default high‑thinking level. Neon’s cheating rate on Terminal‑Bench 4.0 matched Carbon’s 0.0%, with a 1.98% task refusal rate and a 2.45% shortcut rate in an in‑house coding evaluation—the lowest among Gemini 4 models.

Google claims Neon produces more secure code than Carbon on the AutoBaxBench and scores 70.7% on CWE‑bench v2, suggesting improved security outcomes for generated code.

Source details: tbreak.com

Why it matters

Gemini 4 Neon represents Google’s effort to broaden access to high‑capacity, high‑quality generative AI for developers and enterprises that need to handle large codebases or extensive knowledge‑work without paying the premium rates of the top‑tier Carbon model. By offering a 1 million‑token , Neon can process longer files or multi‑file projects in a single request, reducing the need for chunking and simplifying . The pricing structure—especially the steep discount for cached inputs—could make large‑scale code generation and analysis more affordable for businesses, potentially accelerating AI‑assisted software development pipelines. Google’s internal benchmarks suggest Neon delivers strong performance on coding‑specific tests (Terminal‑Bench, FrontierCode) while maintaining lower cheating rates and higher security scores than Carbon, which may influence organizations’ model selection for security‑sensitive code generation. However, the claims are based on Google‑provided metrics; independent verification is pending. Availability outside the U.S. and GCC regions remains unclear, and the model is currently limited to paid Google AI Pro/Ultra subscribers, which may restrict adoption among smaller developers.

The model’s large enables developers to feed entire codebases or extensive documentation into a single request, reducing the overhead of splitting inputs and re‑assembling outputs—a common pain point in current AI‑assisted development workflows.

Lower pricing, especially the heavily discounted cached‑input rate, could make large‑scale code generation and analysis financially viable for mid‑size enterprises and development teams that previously found top‑tier AI costs prohibitive.

Google’s internal benchmark rankings suggest Neon delivers competitive performance on coding‑specific evaluations while maintaining lower cheating and higher security metrics, which is critical for organizations concerned about the reliability and safety of AI‑generated code.

The model’s availability is currently limited to paid Google AI Pro/Ultra subscribers, which may restrict early adoption to larger organizations with existing Google Cloud contracts, but the pricing structure hints at a strategy to attract a broader developer audience over time.

Because the claims are based on Google‑provided data, independent validation will be essential to confirm Neon’s real‑world effectiveness and to compare it against rival offerings from OpenAI’s GPT‑4o, Anthropic’s Claude models, and Microsoft’s MAI Code series.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Interactive Concept Check+10 Points
AI Models Explained Quiz

Which component of an AI application is the machine-learning model itself?

What to watch next

Key indicators to monitor include independent benchmark results that confirm Google’s performance claims, adoption rates among enterprise developers, and any pricing adjustments as the model scales. Watch for announcements about broader regional availability, especially in the GCC, and for feedback from early adopters regarding Neon’s real‑world coding accuracy, security of generated code, and cost‑effectiveness compared with competing models from OpenAI, Anthropic, and Microsoft. Additionally, the rollout of Neon to Google’s Fairwind defender programme may reveal how the model performs in security‑focused contexts.

Independent benchmark publications (e.g., Papers with Code, AI‑Eval) releasing results for Neon on standard coding datasets.

Feedback from early adopters regarding code correctness, security vulnerabilities, and overall developer productivity gains.

Potential expansion of Neon’s availability to additional regions, especially the GCC, and any adjustments to the pricing model.

Updates from Google on the Fairwind defender programme and how Neon performs in security‑focused code generation tasks.

Competitive responses from other AI labs, such as new model releases or pricing changes aimed at the same developer segment.

Related guides & quizzes

Found this useful?