Skip to content
AI.info

The Pulse

Google rolls out Gemini 4 Neon to developers and paid subscribers

Gemini 4 Neon is rolling out through Google's API, business offering and paid consumer products. Google says the model has a one-million-token context window and reports benchmark results across coding, research and cybersecurity.

Google rolls out Gemini 4 Neon to developers and paid subscribers

Gemini 4 Neon is rolling out to developers, businesses and paid subscribers, with Google positioning the model for complex coding, analytical work and scientific tasks. The company says Neon combines a 1 million-token context window with four selectable thinking levels.

In the Gemini app, Neon replaces Argon and Gemini 3.1 Pro is going away, according to 9to5Google. The outlet reports that the model sits between Hydrogen and Carbon in the Gemini 4 family.

Access and pricing

The rollout includes the Gemini API in Google AI Studio and Gemini for business, where the model is available through the API and model dropdown. Access through Google Antigravity is for paid Google AI Pro and Ultra subscribers; the Gemini app rollout is also for Pro and Ultra subscribers.

The model costs $3 per million input tokens, with cached input priced at $0.15 per million tokens, a 95% discount. Output is priced at $15 per million tokens.

Neon has four thinking levels, with high as the default and xhigh intended for deeper, multi-step reasoning. Google says that at the default setting, Neon had a 32% lower cost per task than Carbon across third-party benchmarks. At xhigh, the company says, Neon recorded its highest intelligence score while costing 36%+ less per task than competing models at their highest setting.

Benchmark results

Google reports that Neon ranked third overall on the Artificial Analysis Intelligence Index, with a score of 61 at xhigh. It also reports a second-place score of 70.9% on the Vals Index and first-place results on the Legal Research Bench and Tax Agent Bench.

The company says Neon placed first overall on the Artificial Analysis Coding Agent Index with a score of 58.6 and scored 65.2% on Terminal-Bench 4.1. On FrontierCode 2.0, it scored 46.2% at its default high setting.

For scientific and mathematical reasoning, Google reports scores of 70.5% on Terminal-Bench Science, 84.0% on RiemannBench and 71.3% on HLE-Diamond. The company also says Neon recorded its lowest fabrication rate yet in internal Google Antigravity testing.

Agent workflows and cyber access

In Google Antigravity, Neon can act as an orchestrator directing parallel Hydrogen subagents, as well as work on its own. Google says the model is designed for long-running, tool-heavy tasks and can make targeted edits while debugging and refactoring across repositories.

A separate version, Gemini 4 Neon for cyber, is due to roll out to trusted defenders in the coming days through the Fairwind Program. Google reports that the cyber version scored 76% on the hard tier of AutoBaxBench, which tests code generation. It also reports a score of 70.7% on Collinear’s CWE-bench v2 for patching vulnerabilities, a 3.4% uplift over Carbon.

The company says it paired Neon with real-time safeguards against cyber and biosecurity misuse comparable to those used for Carbon. It also says its internal monitor recorded lower alert rates for chain-of-thought and action misalignment in coding use than the baselines set by Carbon.

Sources

  1. Google
  2. 9to5Google

Companies in this story

Related stories

Explore

More articles