Google has filled the gap in the middle of its Gemini 4 family. Gemini 4 Neon, announced on Wednesday, November 18, is a new mid-tier model built for complex coding, analytical reasoning and production-grade enterprise work — and its price is the story: $3 per million input tokens and $15 per million output tokens in the Gemini API, with cached input at $0.15. Google says Neon runs about three times faster than its strongest model, Carbon, on most tasks, while costing roughly a third less per finished task. It is the third model in a month that has become a price war across the frontier labs.
What Neon is for
Neon sits between the two models Google has already shipped this autumn. Carbon, launched October 28, is the range’s heavyweight for the hardest, longest-running projects; Hydrogen, from November 12, is the fast, cheap everyday engine. Neon is aimed at the work in between: production code, multi-step analysis and enterprise workflows that need more reasoning than Hydrogen offers but do not justify Carbon’s price or pace.
Like its siblings, Neon ships with four configurable thinking levels — low, medium, high and xhigh — so developers choose how much reasoning a task gets and pay only for the depth it needs. Its context window is 1 million tokens, the same as the rest of the family.
The benchmarks: strong on coding, still third overall
The independent picture comes from Artificial Analysis, which runs the same tests for every model. On its Intelligence Index, Neon scores 61 at the xhigh thinking level, ranking third overall — behind Claude Fable 5.2 at 65 and OpenAI’s GPT-6.2 Astra at 62, and ahead of Carbon’s 60. Neon ranks second on the Vals Index at 70.9% and second on the Agent Arena Index at 1,336, but first on the Artificial Analysis Coding Agent Index at 58.6 — the strongest independent result Google has claimed for any of its models.
Google’s own numbers are, as usual, the company’s own: 65.2% on Terminal-Bench 4.1, 46.2% on FrontierCode 2.0, and best-in-class results on the science tests — 70.5% on Terminal-Bench Science, 84.0% on RiemannBench and 71.3% on HLE-Diamond, all at default high thinking. No outside lab had independently reproduced those figures at launch, and they come from benchmarks Google chose.
Neon’s sharpest results are in cybersecurity. Google reports state-of-the-art scores on four of its cyber tests: 76% on the hard tier of AutoBaxBench, 51.26% on WebExploitBench v1.2, 90.5% on SEC-bench Pro and 70.7% on CWE-bench v2. Unlike the general benchmarks, these numbers are aimed at a specific audience: trusted defenders, who will get access through Google’s Fairwind Program in the coming days.
Safety claims
Google also reports that Neon has the lowest fabrication rate yet recorded in internal testing in Google Antigravity, its agent-first development platform; a 0.0% cheating rate on Terminal-Bench 4.0; and a 0.0% attack success rate on the Gray Swan indirect-prompt-injection benchmark at the highest thinking level. All three are internal results, and the safety case will be tested by outside researchers in the weeks after launch.
Price and availability
Neon is available from today, November 18, in the Gemini API in Google AI Studio on the paid tier, in Gemini for business, and in Google Antigravity for paid Pro and Ultra subscribers. Pro and Ultra subscribers can also pick it from the model dropdown in the Gemini app, and tasks inside Gemini can now run on Neon. There is no free API tier at launch — unlike Hydrogen, which launched with one.
The price puts Neon exactly halfway between its siblings on input cost: Hydrogen costs $1.50 per million input tokens, Neon $3, and Carbon $4 at its standard rate. Cached input — the discounted rate for repeated context — is $0.15 per million tokens, a 95% discount that makes long, repeated agent sessions far cheaper than the headline price suggests.
Why it matters
Neon arrives six days after Gemini 4 Hydrogen and three weeks after Gemini 4 Carbon, completing a family that now covers every price point from $1.50 to $4 per million input tokens. The timing is no accident: OpenAI’s GPT-6.2 Astra launched five days after Hydrogen, on November 17, at $8 per million input tokens, and Google’s message is that comparable capability can now be had for less than half that price — Neon is, Google says, 50–80% cheaper per task than the top competing frontier models.
The coding-agent ranking is the more durable signal. Independent tests, not vendor slides, now put a Google model first in coding — a category Anthropic and OpenAI have led for most of the year. If that ranking holds as more evaluations arrive, Neon becomes the default choice for teams that run agents all day and measure cost per finished job rather than cost per token.
What to watch
Three things will decide whether Neon matters beyond launch week: independent cost-per-task figures to test Google’s 32% saving against Carbon, the Fairwind rollout to defenders, and whether developers actually switch — the Gemini app, Antigravity and the API already offer the frontier models, so Neon must win on daily-use quality, not on a benchmark table.
Correction: An earlier version of this article stated that GPT-6.2 Astra launched the day before Gemini 4 Hydrogen. It has been updated to reflect that Astra was announced on November 17, five days after Hydrogen went live on November 12. 19 November 2026
