Gemini 4 Neon
- Provider
- Status
- Current
- Context
- 1,000,000 tok
- Price
- $3 / $15 /MTok
- Knowledge
- 2026-03
Gemini 4 Neon is a Google model for complex coding, analysis and business work, released on 18 November 2026. Google places it between Gemini 4 Carbon, which it describes as its model for the hardest, long-horizon projects, and Gemini 4 Hydrogen, its model for fast everyday work, and says Neon “balances depth and efficiency” (Google). Its model card says it is adapted from Carbon.
Artificial Analysis scored it 61 on the Intelligence Index v5.0.1 at its highest thinking setting, xhigh, at launch, a point above Gemini 4 Carbon. It costs $3 per million input tokens and $15 per million output tokens on the paid Gemini API, and it is in the Gemini app for Google AI Pro and Ultra subscribers.
Quick specs
| Gemini 4 Neon | |
|---|---|
| Released | 18 November 2026 |
| Availability | Paid Gemini API, Gemini for business, Antigravity and the Gemini app for Google AI Pro and Ultra |
| Input price | $3.00 / MTok |
| Cached input | $0.15 / MTok (95% off) |
| Output price | $15.00 / MTok |
| Batch | $1.50 / $7.50 per MTok |
| Context window | 1,000,000 tokens, input and output combined, one price throughout |
| Max output | 64,000 tokens by default; up to 1,000,000 through background long decode |
| Thinking levels | Low, medium, high (default), xhigh |
| Input | Text, images, video, audio |
| Knowledge cutoff | March 2026 |
| AA Intelligence Index | 61 at launch (v5.0.1, xhigh) |
| Cost per AA index task | $6.44 |
| API model ID | gemini-4-neon |
Who can use it
Paying users, from 18 November 2026. Google began rolling it out on launch day (Google):
- Developers: on the paid tier of the Gemini API in Google AI Studio, and in Android Studio. It is not on the free tier (Google AI Studio pricing).
- Businesses: in Gemini for business, through the API or the model menu.
- Antigravity: for paid Google AI Pro and Ultra subscribers. Google says Neon can act as the Orchestrator there, directing Gemini 4 Hydrogen sub-agents working in parallel.
- Gemini app: in the model menu for Google AI Pro and Ultra subscribers.
Gemini 4 Neon for cyber, a version for security work, is going to trusted defenders through Google’s Fairwind Program, the vetted-access scheme Google also used for Carbon and Gemini 4 Argon.
Pricing
Google charges $3 per million input tokens and $15 per million output tokens, thinking tokens included, flat across the full 1 million-token context window. Cached input costs $0.15, a 95% discount. Batch and Flex requests cost half, $1.50 and $7.50, and Priority requests cost $5.40 and $27 (Google AI Studio pricing).
That is below Gemini 4 Carbon’s $4/$20 standard price, which applies from 1 December 2026 when its introductory $2/$10 ends, and twice Gemini 4 Hydrogen’s $1.50/$7.50.
Benchmark performance
Independent
| At Neon’s launch (18 Nov 2026) | Gemini 4 Neon | GPT-6.2 Astra | Gemini 4 Carbon | Claude Opus 5.5 | GPT-6.2 Sol |
|---|---|---|---|---|---|
| Intelligence Index v5.0.1 | 61 (xhigh) | 62 | 60 | 59 | 57 |
| Cost per index task | $6.44 | $4.00 | $7.76 | $10.04 | $1.01 |
| Coding Agent Index v2.1 | 59 | n/a | 55 | 58 | n/a |
| Price (in/out per MTok) | $3 / $15 | $8 / $40 | $2 / $10 intro, then $4 / $20 | $4 / $20 | $2 / $10 |
Source: Artificial Analysis for Neon, and for Carbon’s and Opus 5.5’s Coding Agent Index scores. The other figures are Artificial Analysis’s readings as carried on our ranking on the day Neon launched; Carbon’s cost per task is at its $4/$20 standard price. For the Coding Agent Index, Artificial Analysis ran Neon in Google’s Antigravity CLI and Opus 5.5 in Claude Code.
Intelligence. At launch, 61 on the index at xhigh, which Artificial Analysis called third overall, behind GPT-6.2 Astra and Claude Fable 5.2. It scored 60 at the default high setting, 55 at medium and 44 at low.
Cost per task. $6.44 per completed index task at xhigh, about 17% less than Carbon at Carbon’s standard price. Artificial Analysis found Neon used about 15% more output tokens per task than Carbon, so the saving comes from its lower token price. At the high setting, it says Neon matched Carbon’s score at less than half the cost per task. Running the whole index at xhigh took 283 million output tokens, against a median of 65.1 million.
Coding. At launch, 59 on Artificial Analysis’s Coding Agent Index v2.1 with the Antigravity CLI, which Artificial Analysis put 3 points ahead of Carbon (55) and in line with Claude Code running Claude Opus 5.5 (58) or Claude Fable 5.2 (58).
Speed. At launch, at xhigh, Artificial Analysis measured 110 output tokens a second, against about 57 for Carbon, with 56.51 seconds to the first answer token and about 21 minutes per index task, against about 36 for Carbon.
Knowledge work. At launch it scored 1,700 Elo on Artificial Analysis’s AA-Briefcase at xhigh.
Google’s own figures
| Benchmark | Gemini 4 Neon | Claude Opus 5.5 | GPT-6.2 Sol |
|---|---|---|---|
| Vals Index v2.2 | 70.9% | 70.1% | 66.1% |
| CursorBench 4.1 | 65.4% | 58.1% | n/a |
| Terminal-Bench 4.1 | 65.2% | 65.8% | 61.5% |
| FrontierCode 2.0 | 46.2% | 47.3% | 43.8% |
| SWE-sweep (each lab’s own harness) | 39.0% | 10.1% | 8.3% |
| SWE-sweep (mini-SWE harness) | 12.8% | 3.6% | 11.6% |
| AutomationBench | 56.3% | 42.5% | 41.3% |
| Agents’ Last Exam v1.1 | 75.8% | 71.6% | 73.0% |
| Terminal-Bench-Science 0.1 | 70.5% | 63.3% | 68.0% |
| HLE-Diamond | 71.3% | 67.9% | 64.0% |
| Chartography | 73.5% | 66.3% | 75.1% |
| GraphWalks (256K to 1M tokens) | 95.2% | 66.8% | 80.1% |
| CWE-Bench v2 | 70.7% | 67.1% | 66.0% |
| SEC-bench Pro | 90.5% | n/a | 88.3% |
Source: Google’s model card, vendor-reported, as of November 2026. On Google’s table, Opus 5.5 leads Neon on Terminal-Bench 4.1 and FrontierCode 2.0, and GPT-6.2 Sol leads it on Chartography. Google also says Neon has 32% lower cost per task than Carbon at the default high setting, measured on Artificial Analysis’s index; Artificial Analysis’s own comparison at that setting, using Carbon’s $4/$20 standard price, found less than half the cost per task. It published no SWE-bench Pro or SWE-bench Verified figure.
Safety. Google did not run its full set of frontier safety tests on Neon, because its capability check against Argon fell below the threshold that requires them; it carries forward Argon’s results, under which Google says it cannot rule out that the model reaches its critical capability levels for biological and cyber misuse (model card). Google says its full safety and security report for Neon will follow.
How it compares
vs Gemini 4 Carbon
Neon is adapted from Gemini 4 Carbon, and Google keeps Carbon as its model for the hardest, longest projects. When Neon launched, Artificial Analysis scored Carbon 60 on the v5.0.1 index, a point lower, and 55 on its Coding Agent Index, which Artificial Analysis put 3 points behind Neon. A completed index task cost $7.76 for Carbon at its $4/$20 standard price, against Neon’s $6.44. Until 30 November 2026 Carbon’s introductory price is $2/$10, below Neon’s.
vs Claude Opus 5.5
When Neon launched, Claude Opus 5.5 scored 59 on the v5.0.1 index, two points lower, at $4/$20 and $10.04 per completed task. On Artificial Analysis’s Coding Agent Index at Neon’s launch, Claude Code with Opus 5.5 scored 58 to Neon’s 59, which Artificial Analysis called in line. On Google’s own table Opus 5.5 is ahead on Terminal-Bench 4.1 and FrontierCode 2.0, by under a point and a half on each.
vs GPT-6.2 Astra
When Neon launched, GPT-6.2 Astra scored 62 on the v5.0.1 index, a point higher. Its tokens cost more, at $8/$40, but a completed task cost less, $4.00 against Neon’s $6.44.
vs GPT-6.2 Sol
When Neon launched, GPT-6.2 Sol scored 57 on the v5.0.1 index, four points lower, at $2/$10 and $1.01 per completed task. On Google’s table Neon leads Sol on every shared row except Chartography.
Known limitations
Paid only. It is not on the free Gemini API tier, and in the Gemini app and Antigravity it is limited to Google AI Pro and Ultra subscribers.
Verbose at xhigh. It used 283 million output tokens to run Artificial Analysis’s index at xhigh, more than four times the median.
Slow to start at xhigh. Artificial Analysis measured 56.51 seconds to the first answer token at that setting, against 40.64 seconds at high.
No SWE-bench figure. Google published neither SWE-bench Pro nor SWE-bench Verified.
Frequently asked questions
When was Gemini 4 Neon released?
On 18 November 2026. Google began rolling it out that day on the paid Gemini API, in Gemini for business, in Antigravity and in the Gemini app for Google AI Pro and Ultra subscribers.
How much does Gemini 4 Neon cost?
$3 per million input tokens and $15 per million output tokens on the Gemini API, the same across its full 1 million-token context window, with cached input at $0.15. Batch requests cost half.
Can I use Gemini 4 Neon for free?
No. It is not on the free tier of the Gemini API, and in the Gemini app it is for Google AI Pro and Ultra subscribers.
How is Gemini 4 Neon different from Gemini 4 Carbon?
Google’s model card says Neon is adapted from Carbon, and Google describes Carbon as its model for the hardest, long-horizon projects and Neon as a balance of depth and efficiency. At Neon’s launch, Artificial Analysis scored Neon a point higher on its Intelligence Index (61 to 60) and 3 points higher on its Coding Agent Index (59 to 55, as rounded), and Neon was about 17% cheaper per completed task than Carbon at Carbon’s standard price.
Is Gemini 4 Neon better than Claude Opus 5.5?
At Neon’s launch, Artificial Analysis scored it 61 to Opus 5.5’s 59 on its Intelligence Index, and 59 to 58 on its Coding Agent Index, which Artificial Analysis called in line, and a completed index task cost $6.44 against $10.04. On Google’s own table, Opus 5.5 is slightly ahead on Terminal-Bench 4.1 and FrontierCode 2.0.