Drafted with AI tools and edited by a human. The figures and conclusions were checked by a GPUniq editor.
TL;DR
Gemini 4 Neon costs $3.00 per million input tokens and $15.00 per million output tokens. It uses a 1,048,576-token unified context window and scored 61.4 on the Artificial Analysis Intelligence Index. The model targets high-performance reasoning in software engineering and science.
Gemini 4 Neon Pricing
Gemini 4 Neon costs $3.00 per million input tokens. Output is $15.00. Google DeepMind released the model on November 18, 2026, for paid platforms like Gemini for Business and Google AI Studio. You won't find this on the free tier. Full access requires a subscription to Google Antigravity or Gemini Pro and Ultra.
| Metric | Rate ($/1M tokens) |
|---|---|
| Input Tokens | 3.00 |
| Cached Input Tokens | 0.15 |
| Output Tokens | 15.00 |
Cached Token Rates
Cached tokens are cheap. They cost $0.15 per million. If you're running long chat sessions or sending the same codebase repeatedly, this saves money. It's built for production environments where context reuse is high. Enterprise users benefit most when handling large-scale data retrieval.
Context Window and Output Limits
The model supports a unified context window of 1,048,576 tokens. This handles text, image, audio, and video. While the input window is massive, synchronous output is capped at 65,536 tokens. For longer responses, you'll need the background long decode feature. It maintains high retrieval accuracy across the entire space.
Reasoning and Intelligence Benchmarks
Neon scored 61.4 on the Artificial Analysis Intelligence Index for xhigh thinking. On the Vals Index, it hit 70.9%. That puts it slightly behind Claude Fable 5.2. You can choose between four thinking levels: low, medium, high, and xhigh. This lets you trade speed for depth depending on the task.
Comparison to Gemini 4 Carbon
Neon offers quality similar to Gemini 4 Carbon but with lower latency and cost. It scored 56.3% on AutomationBench 1.0.6. That beats the 53.9% score from Gemini 4 Carbon at High effort. It's a better choice for software engineering and scientific work.
Scientific and Technical Performance
The model handles math and science well. It reached 84.0% on RiemannBench and 71.3% on HLE-Diamond. It also scored 70.5% on Terminal-Bench Science. These numbers suggest it is a reliable tool for researchers who need to process complex technical data.
| Benchmark | Score (%) |
|---|---|
| RiemannBench | 84.0 |
| HLE-Diamond | 71.3 |
| Terminal-Bench Science | 70.5 |
| Vals Index | 70.9 |
Automation and Tool Calling
Automation works via JSON schema and tool calling. The 56.3% pass rate on AutomationBench 1.0.6 shows it handles multi-step tasks. Hallucinations are lower than older models, but they still happen. You can use it for logic where the model must interact with external systems.
{
"model": "gemini-4-neon",
"task": "workflow_automation",
"thinking_level": "xhigh",
"tools": [
{
"name": "api_connector",
"parameters": {
"action": "sync_data"
}
}
]
}
Related on the GPUniq blog
Methodology
Facts are compiled from the vendor's own announcement and documentation, linked below, at the time of writing; where the vendor hasn't published a number it is marked as such. GPUniq prices, when the model is already available here, come from our live catalog.
Sources
- 1.Introducing Gemini 4 Neon — Google (accessed )
- 2.Gemini 4 Neon — Model Card — Google DeepMind (accessed )
- 3.Gemini 4 Neon - API Pricing & Providers | OpenRouter — OpenRouter (accessed )
- 4.Gemini 4 Neon AutomationBench Results: 56.3% | Zapier — Zapier (accessed )
- 5.Benchmark Results & Review for Gemini 4 Neon — SpringPrompt (accessed )
- 6.Gemini 4 Neon Model Evaluation — Google DeepMind (accessed )
Want cheap GPUs for your next project?
Browse live GPU prices and rent the right card in seconds — H100, A100, RTX 4090, and 50+ more models.
