Gemini 4 Neon is third on raw intelligence, first on coding agents and cheaper per token than either rival's flagship. Anthropic still has the single strongest model.
For two years the AI race looked like Anthropic and OpenAI trading first place while Google played catch-up. Gemini 4 Neon, released on November 18, 2026, changes that picture. It lands one point behind OpenAI's GPT-6.2 Astra on the leading independent index, tops the coding-agent rankings, and costs $3/$15 per million tokens against $8/$40 for both rivals' flagships.
Our verdict: Google has caught up, and on price it's ahead. It hasn't passed Anthropic. Claude Fable 5.2 is still the strongest single model, by four points.
Has Google caught up with OpenAI and Anthropic?
Mostly, yes. Here is where Google's best model lands on each leaderboard that matters, checked on launch day.
| Leaderboard | Leader | Gemini 4 Neon |
|---|---|---|
| Intelligence Index Artificial Analysis | Fable 5.2, 65 | 3rd, 61 |
| Coding Agent Index Artificial Analysis | Neon, 58.6 | 1st |
| Vals Index real-world tasks | Fable 5.2, 71.3% | 2nd, 70.9% |
| Agent Arena 4,200 real sessions | A rival model | 2nd, 1,336 |
| Text Arena people vote on answers | 3rd, 1,518 |
Sources: Artificial Analysis, Vals AI, LMArena, as reported by TestingCatalog, November 18, 2026. Google models hold all three top Text Arena spots.
The pattern is clear. On blind votes and coding agents, Google leads or sits second. On the hardest reasoning tests, Anthropic and OpenAI still have the edge, though Neon is within a point of Astra.
What is Gemini 4 Neon?
Think of it as Google's new middle manager who turns out to be sharper than the boss. It sits between Gemini 4 Carbon, the big flagship, and Hydrogen, the fast cheap one, yet it outscores Carbon 61 to 60.
Google says Neon costs 32% less per task than Carbon at its default setting, and that it refuses far fewer normal requests while keeping Carbon's safety protections. In Antigravity, Google's coding app, it can direct a team of Hydrogen helpers working in parallel.
Gemini 4 Neon price and how to get it
- Included in Google AI Pro and Ultra
- Now powers Tasks and deeper analysis
- Works inside Gmail, Docs and Drive
The catch: free Gemini users don't get Neon.
- Paid tier in Google AI Studio
- Batch and Flex at half price
- 1 million token context
The catch: not on the free API tier.
- Google's agent coding app
- Talk to it out loud with voice mode
- Where the coding-agent scores come from
The catch: needs a paid Google plan.
The standout feature: you can talk to it while it codes
Antigravity has a voice mode, built on Google's Gemini Live. You speak, Neon answers out loud, and it keeps reading code, editing files and running commands while you talk. It feels like pair programming with a colleague on a call. We tried it on our Google AI Pro account, and it's the feature we'd show a friend first.
Why it matters: you can explain a bug the way you'd explain it to a person, and steer the work without stopping to type.
The limits: deleting files or force-pushing still needs an on-screen click, and your audio goes to Google's servers. It's in the Antigravity app and command-line tool, announced on November 6. Codex also talks back, while Claude Code only takes dictation: see Antigravity vs Codex vs Claude Code. Not sure which Google app to download? Read Gemini app vs Antigravity.
Gemini 4 Neon benchmarks
Google picked weaker rivals for its own chart, so read it as a ceiling. The independent numbers in the leaderboard table above are the fairer test, and they still put Neon in the top three.
Gemini 4 Neon vs Claude Fable 5.2 vs GPT-6.2 Astra
| Neon | Fable 5.2 | Astra | |
|---|---|---|---|
| Intelligence score | 61 | 65 | 62 |
| Per million in / out | $3 / $15 | $8 / $40 | $8 / $40 |
| Cost per test task | $6.44 | $26.54 | $4.00 |
| Speed, tokens a second | 110 | 112 | 66 |
| In a $20 plan | ✓ | ✓ | Work mode |
| Image creation in the app | ✓ | ✗ | ✓ |
Scores, cost per task and speed from Artificial Analysis, checked November 18, 2026. Prices from Google, Anthropic and OpenAI.
The catch with Neon is how much it writes. Its tasks took 21.5 minutes against a 12.4-minute median for similar models, which is why its cost per task sits above Astra's despite the lower token price. For a full head-to-head with the mid-tier rivals, see Gemini 4 Neon vs Claude Opus 5.5 vs GPT-6.2 Sol.
What early users say
Praise: AI tester Chubby had it build 17 interactive demos in Antigravity in a day, from shooters to 3D worlds (thread). Hacker News commenters rate Gemini 4's writing as the most natural-sounding of the big models.
Complaints: the same tester's first build stopped on a server error and needed a second pass. On Hacker News, several people suspect the scores are tuned for benchmarks, and many find Google's element names (Carbon, Neon, Hydrogen, Argon) impossible to rank.
One developer on DEV Community argues the middle tier is where daily coding happens: refactors, tests, reviews and bug fixes. That matches the numbers. Neon's strength is finishing that work for less, while Fable still owns the hardest problems.
We tested Gemini 4 Neon on real work
We gave Neon the same three tasks we gave Claude Fable 5.2, Opus 5.5, GPT-6.2 Astra and Sol, in the Gemini app on Google AI Pro, with thinking set to High and a fresh temporary chat each time.
Twelve tasks, every deadline and dependency met, 6.5 hours to spare. It added a travel buffer before the dentist and offered a second layout with a warning about its trade-off.
A $2.46 loss, $44.16 and $15.54, all exact, with the same card-fee rounding note Fable and Astra gave.
Spotted low iron stores without anemia and explained it as a checking and a savings account. It showed no sources and listed no warning signs.
What we found: Neon matched Fable 5.2 on care, flagging its assumptions and offering alternatives, and gave the easiest health explanation of the five. It was also the slowest every time, taking one to three minutes per answer. Full results for all five models are in our best AI model guide.
Who should switch to Gemini 4 Neon?
Neon in Google AI Pro is the strongest model you can get inside Gmail and Docs, and it tops the coding-agent index at well under half Fable's token price. Fable 5.2 still leads the independent index by four points. Astra did the Artificial Analysis tasks for $4.00 against Neon's $6.44, and Sol for $1.01.
The verdict
Gemini 4 Neon puts Google back in the top three, and for many buyers it's the best deal of the three. It's a point behind GPT-6.2 Astra on raw intelligence, first on coding agents, and much cheaper per token. Anthropic's Claude Fable 5.2 remains the strongest model you can buy. If you already pay for Google AI Pro, you're now getting a frontier model for $19.99.
FAQ
What is Gemini 4 Neon?
Gemini 4 Neon is Google's mid-tier AI model, released on November 18, 2026. It sits between Gemini 4 Carbon and Gemini 4 Hydrogen, costs $3/$15 per million tokens, reads up to 1 million tokens, and scores 61 on the Artificial Analysis Intelligence Index, one point above Carbon.
Is Gemini 4 Neon good?
Yes. It ranks third on the Artificial Analysis Intelligence Index behind Claude Fable 5.2 and GPT-6.2 Astra, first on the Artificial Analysis Coding Agent Index at 58.6, and second on the Vals Index at 70.9%. In our own tests it passed all three everyday tasks we gave it, though it was the slowest of the five models we tried.
Has Google caught up with OpenAI and Anthropic?
Nearly. Gemini 4 Neon scores 61 on the Artificial Analysis Intelligence Index against 62 for OpenAI's GPT-6.2 Astra and 65 for Anthropic's Claude Fable 5.2. Google leads on coding agents and price per token, and its models hold the top three Text Arena spots.
How much does Gemini 4 Neon cost?
$3 per million input tokens and $15 per million output tokens on the Gemini API, with cached input at $0.15. In the Gemini app it comes with Google AI Pro at $19.99 a month and Google AI Ultra.
Is Gemini 4 Neon free?
No. It isn't on the free Gemini API tier, and in the Gemini app it needs a Google AI Pro or Ultra subscription. Free Gemini users get Google's smaller models.
Can you talk to Gemini 4 Neon out loud?
Yes. Antigravity, Google's coding app, has a voice mode built on Gemini Live. You speak, Neon answers out loud and keeps working on your code while you talk, and you can interrupt it or approve steps by voice. Risky actions like deleting files still need an on-screen click.
Is Gemini 4 Neon better than Claude?
Better than Claude Opus 5.5 on most tests, with 61 points against 59 and a lower price. Not better than Claude Fable 5.2, which scores 65 and leads the Vals Index by 0.4 points, though Fable costs more than twice as much per token.
Is Gemini 4 Neon better than Gemini 4 Carbon?
On the Artificial Analysis Intelligence Index, slightly: 61 against 60, at about 17% lower cost per task. Google still positions Carbon for the hardest long-running projects and Neon for everyday heavy work.