Gemini 4 Neon
Google's mid-tier: 61 points puts it third overall, one point above Carbon at 50% higher price
Benchmark scores
How it is measured →Ranked 3 · bars are scaled to the field maximum for that benchmark.
| Benchmarks | Score | Vs. best in field | Gap |
|---|---|---|---|
| AA Intelligence Index v5 Intelligence index | 61.0 | 94% | −4.0 pts |
| AA Intelligence Index v4.3.2 | — | — | not indexed |
| GPQA Diamond | — | — | not indexed |
| Terminal-Bench 4.0 | — | — | not indexed |
| LiveBench Design | — | — | not indexed |
| Vals Index Vals composite index | 70.9 | 97% | −2.2 pts |
Assessment
Released on 18 November 2026, the third new tier in the Gemini 4 family in 21 days (Carbon on 10-28, Hydrogen on 11-12, Neon on 11-18), officially pitched between Carbon (the hardest long-horizon projects) and Hydrogen (fast everyday execution). AA Intelligence Index v5.0.1 lists four effort tiers: xhigh 61 / high 60 / medium 55 / low 44. This site records the top tier at 61, third overall, above Carbon's top setting of 60 and behind only Claude Fable 5.2 at 65 and GPT-6.2 Astra at 62; Google's own launch post also says "#3 overall, 61 on the AA Intelligence Index at xhigh", so the two sources agree. Pricing is $3 input / $15 output with cache hits at $0.15 (a 95% discount) and a 1M context, taken from Google's official launch post - the most expensive of the four Gemini 4 tiers (Carbon and Argon sit at the introductory $2/$10, Hydrogen at $1.5/$7.5), 50% above Carbon. Availability follows the official line: the Gemini API in AI Studio, Gemini for business and the Pro/Ultra tiers of Antigravity and the Gemini app from day one; Hydrogen's "available to everyone" does not extend here, and the free tier does not get it; a cyber variant for defenders opens through the Fairwind programme later. It scores 70.9% on the Vals Index, which Google ranks second. Speed and latency are 110 tok/s and 56.51 seconds to first answer token per AA; the trade-off is slow and verbose - 21.5 minutes and $6.44 per index task, with 283M output tokens for the full suite against a median of 65.1M. Three columns are left blank with the reason noted: Terminal-Bench 4.0 here is AA's own run, and AA's model page could not be reached from either the local machine or the server this time (TCP timeouts), so the embedded record was unavailable, while Google's post only reports Terminal-Bench Science at 70.5% and HLE-Diamond at 71.3% - different evaluations that are not mixed in; GPQA and LiveBench have no first-party figures either, and the parameter count is unpublished, so pt/pa stay empty.
At the same price
The four models closest on blended price, side by side.
| Model | v5 | Blended | tok/s | Score gap |
|---|---|---|---|---|
| Claude Opus 5.5Anthropic | 59.1 | $8.00 | 116 | -1.9 |
| Claude Sonnet 5.5Anthropic | 54.5 | $4.00 | 146 | -6.5 |
| GPT-6.2 SolOpenAI | 57.0 | $4.00 | 85 | -4.0 |
| Gemini 4 CarbonGoogle | 60.0 | $4.00 | 80 | -1.0 |
Compare side by side
Line it up against other models on price, speed and every benchmark.
The comparison tool lines up to 4 models side by side with switchable normalisation — built for purchasing decisions.
Open the comparison tool → Back to leaderboardRelated benchmarks
Data sources & methodology
Full methodology & sources →Every number on this page comes from published evaluations. We run no tests of our own and accept no vendor-submitted scores. Main sources:
Evaluation data in the wider industry also comes from public datasets, leaderboard platforms and their evaluation harnesses, private in-house sets, production logs and user feedback, and synthetic data. This site's current rankings use only the parts that are public, independently checkable and clearly defined.