Gemini 4 Neon on BenchGecko · price and context
Google's Gemini 4 Neon lists at $3 input and $15 output per million tokens with a 1,048,576 token context · 2 times Hydrogen per token and 25% below Carbon and Argon once their list price starts on December 1, 2026.
What was added
Google DeepMind released Gemini 4 Neon on November 18, 2026. It is now tracked on BenchGecko under Google DeepMind, together with its batch row. The API id is gemini-4-neon. Google positions Neon for fast end to end task completion across software engineering, knowledge work and science workflows.
Context and limits
Neon accepts text, image, video, audio and PDF input and returns text. Google lists a unified context window of 1,048,576 tokens for input and output combined. The default synchronous output limit is 65,536 tokens, and long decode raises output up to 1,048,576 tokens. The model supports thinking with four levels, function calling, structured outputs, code execution, computer use and grounding with Google Search and Google Maps. It is not available on the free tier.
Price on BenchGecko
- Gemini 4 Neon · $3 input · $15 output · $0.15 cached input per million tokens
- Gemini 4 Neon (batch) · $1.50 input · $7.50 output · $0.075 cached input per million tokens
Cached input is 95% below the input price, and caching is implicit only. Output pricing includes thinking tokens.
How it compares on price
Gemini 4 Hydrogen lists at $1.50 input and $7.50 output, so Neon costs 2 times as much per token. Gemini 4 Carbon and Gemini 4 Argon list at an introductory $2 input and $10 output through November 30, 2026, which makes Neon 50% more expensive per token until then. From December 1, 2026 Carbon and Argon move to $4 input and $20 output, and Neon becomes 25% cheaper per token than both. All four share the same 1,048,576 token context window.
Google also reports cost per task rather than per token: at its default high thinking level, Neon completes tasks at 32% lower cost than Carbon.
Vendor benchmarks
Google reports these figures in its announcement:
- Artificial Analysis Intelligence Index · 61 at xhigh thinking, ranked 3rd overall
- Artificial Analysis Coding Agent Index · 58.6, ranked 1st
- Vals Index · 70.9%, ranked 2nd
- Agent Arena Index · 1336, ranked 2nd
- Terminal-Bench 4.1 · 65.2%
- Terminal-Bench Science · 70.5%
- FrontierCode 2.0 · 46.2% at the default high setting
- Agents' Last Exam v1.1 · 75.8%
- HLE-Diamond · 71.3% · RiemannBench · 84.0%
These are self reported results. BenchGecko has no independent scores for Neon yet and will add them to the model page as publishers report them.
Sources
Source: BenchGecko
Written automatically from BenchGecko's tracked data for this period; every number comes from the changelog, model, mindshare and pricing datasets. Checked against the source data before publishing.