The new model costs $3 per million input tokens and $15 for output, and independent testers put it behind only Anthropic's and OpenAI's top models.
· 4 min read
Google released Gemini 4 Neon on Wednesday, a mid-tier model it says delivers frontier-level reasoning for coding, knowledge work and science at $3 per million input tokens and $15 per million output tokens. Independent testers rank it third on a widely followed intelligence index, behind only the top models from Anthropic and OpenAI.
Where Neon sits in the Gemini 4 family
Neon follows a run of Gemini 4 releases that began with Argon on September 30. In its announcement, Google casts Carbon as the tool for the hardest long-horizon projects and Hydrogen as the fast everyday option, with Neon balancing depth and efficiency between them. It arrives about a week after Hydrogen, and Artificial Analysis counted it as Google's fourth Gemini 4 model in seven weeks.
Availability and pricing
Neon is rolling out in the Gemini API through Google AI Studio and in Gemini for business. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity, and Pro and Ultra subscribers can pick it in the Gemini app. Google lists a one-million-token context window and a cached-input rate of $0.15 per million tokens.
SiliconANGLE placed those rates against rivals. OpenAI's GPT-6.2 Sol costs $2 per million input tokens and $10 for output, while Anthropic's Claude Opus 5.5 costs $4 and $20, so Neon lands between them. The publication added that many organisations may decide Neon's edge on some workloads does not justify paying more than they would for Sol.
What the benchmarks say
Google says Neon scores 61 on the Artificial Analysis Intelligence Index at its highest thinking setting, third overall. SiliconANGLE, which checked the independent index, reported that Anthropic's Claude Fable 5.2 leads with 65 and OpenAI's GPT-6.2 Astra follows with 62. Google also cites a second-place finish on the Vals Index at 70.9 percent and a score of 1,336 on Arena's new Agent Arena Index, drawn from more than 4,200 real-world agent sessions.
For coding, Google reports 65.2 percent on Terminal-Bench 4.1 and a first-place 58.6 on the Artificial Analysis Coding Agent Index. On SWE-sweep it reports 12.8 percent with the mini-SWE harness and 39.0 percent inside Antigravity, a gap Google credits to its own tooling. These numbers come from Google's blog post, not from independent replication.
Enterprise and science claims
Google pitches Neon at spreadsheet, PDF, chart, audio and video work inside the single million-token window. It reports first place on the Legal Research Bench and Tax Agent Bench, 56.3 percent on Zapier's AutomationBench and $16,843 on Andon Labs' Vending-Bench 2, a long-horizon business simulation. On science and maths it claims the best results on Terminal-Bench Science at 70.5 percent, RiemannBench at 84.0 percent and HLE-Diamond at 71.3 percent.
Speed and cost claims differ by setting
SiliconANGLE's reading of Google's materials is that Neon finishes most tasks about three times faster than Carbon at roughly 60 percent lower cost. Google's blog is narrower: at the default high thinking level, it says Neon costs 32 percent less per task than Carbon on the third-party index. Artificial Analysis, which tested the xhigh setting behind the score of 61, put the saving at about 17 percent for a small gain in intelligence over Carbon. The figures differ because they describe different thinking levels, so buyers should test their own workloads before relying on either number.
Safety and the cyber variant
Google says Neon keeps Carbon-level protections against severe misuse and prompt injection. In its testing the model refused 1.98 percent of tasks on Terminal-Bench 4.0, showed 0.0 percent cheating on that benchmark and took shortcuts on 2.45 percent of tasks in its coding cheating evaluation. A separate Gemini 4 Neon for cyber variant goes to eligible defenders in the Fairwind Program over the coming days. Google reports 51.26 percent on WebExploitBench v1.2 and 90.5 percent on SEC-bench Pro for that version, and it has published a model card.
Tasks in Gemini gets a faster option
In the Gemini app, Google says Tasks can now run on Neon, with the aim of finishing multi-step jobs faster than larger models while stretching usage limits. Gadgets365 covered the wider Tasks rollout, including its free availability in India, in a separate report.
What to watch next
The pricing contest is now the clearest test of the release. With Sol below it and Opus above it, Neon gives enterprise buyers a third mid-tier price point, and its case rests on the cost-per-task numbers holding up outside Google's own harness. Independent leaderboards will update as more testers run Neon at each thinking level, and the cyber variant's rollout to Fairwind members should show how the security claims hold in practice. The week's other major AI story, Anthropic's stock market debut, is covered in our report on the first day of trading.
Published November 20, 2026. Gadgets365 will update this article as more information becomes available.
Editor
Pravin KankanePravin Kankane is the editor of Gadgets365, an independent technology publication based in Pune, India. He covers smartphones, AI, Apple, Android and consumer tech launches in India.
