
Gemini 4 Neon Beats Opus 5.5 on the Artificial Analysis Board: Cheaper, Less Hallucination — But Not Everywhere
Vũ Khôi
Tác giả tại Wotech Electronics Blog

Gemini 4 Neon Beats Opus 5.5 on the Artificial Analysis Board: Cheaper, Less Hallucination — But Not Everywhere
Neon launched this morning, replacing Argon — and according to a post by Đỗ Anh Việt in the Nghiện Claude group (19/11/2026), Gemini 4 Neon beats Claude Opus 5.5 on most fronts on the Artificial Analysis leaderboard. Neon scores 61 on the Artificial Analysis Intelligence Index, ahead of Opus 5.5 Max (59); long-context reading hits 95.2% versus Opus 5.5's 66.8%; it costs nearly 2.7 times less; and it takes the global #1 spot on the Coding Agent Index. But the very table in the post shows the picture is not one-sided: Opus 5.5 still wins some tests, and Claude still leads on the agentic knowledge test. All figures, views, and recommendations in this article follow the post and have not been independently verified by Wotech — and the benchmark table in the post's image cites its methodology from Google's eval page (deepmind.google), i.e. figures published by Google itself under its own methodology.
Quick answer: According to the post, Gemini 4 Neon scores 61 (xhigh) and 60 (high) on the Artificial Analysis Intelligence Index, ahead of Opus 5.5 Max (59); long-context reading (GraphWalks 256k–1M tokens) reaches 95.2%, far ahead of Opus 5.5's 66.8%. On cost, Neon (high) costs only $3.74/task for 60 points, nearly 2.7 times cheaper than Opus 5.5 Max ($10.04 for 59 points); the medium tier costs $1.90 for 55 points, over 5 times cheaper than Sonnet 5.5 Max at the same score. Antigravity CLI + Gemini 4 Neon scores 59 on the Coding Agent Index, slightly ahead of Claude Code and Opus 5.5 (both 58) — the first time in a while a Gemini model has overtaken Claude for the #1 spot, per the post. The non-hallucination rate of the Gemini 4 line tops the chart (Hydrogen 91%–92%, Neon 81%–86%), while Opus 5.5 manages only 33%–50%. However, on the Agentic Knowledge test (AA-Briefcase, Elo scale), Claude still leads: Opus 5.5 Max scores 1800, Sonnet 5.5 Max 1815, while Neon xhigh stops at 1700; and the author notes Claude's subsidy is about 3 times better at the same cost as Gemini, so using Gemini now still burns Usage faster than Claude.
The Artificial Analysis Intelligence Index is Artificial Analysis's composite index — per the post, Neon scores 61 (xhigh) on it, ahead of Opus 5.5 Max (59).
Where does Neon win?
According to the post, after Google reached RSI (the AI's ability to improve its own model — as the author puts it), Google's model release pace accelerated sharply, all strong models; Neon launched on the morning of 19/11/2026, replacing the earlier Argon. The highlights from Google and Artificial Analysis:
| Category | Gemini 4 Neon | Claude Opus 5.5 | Note per the post |
|---|---|---|---|
| Intelligence Index | 61 (xhigh), 60 (high) | 59 (Max) | Neon ahead of Opus 5.5 Max |
| Long-context reading (GraphWalks 256k–1M) | 95.2% | 66.8% | Neon far ahead |
| Cost per task | $3.74 (high, 60 pts); $1.90 (medium, 55 pts) | $10.04 (Max, 59 pts) | Neon high nearly 2.7× cheaper; medium over 5× cheaper than Sonnet 5.5 Max at the same score |
| Coding Agent Index | 59 (Antigravity CLI + Neon) | 58 (Claude Code + Opus 5.5) | First time Gemini takes the #1 spot over Claude in a while |
| CursorBench 4.1 | 65.4% | 58.1% | Neon wins |
| Non-Hallucination | Hydrogen 91%–92%, Neon 81%–86% | Opus 5.5 only 33%–50% | Gemini 4 line takes the whole top |
| AutomationBench-AA (SaaS automation) | 78%–79% | 61%–70% | Neon leads |
The table in the post's image also shows Neon winning big on SWE-sweep (provider harness 39.0% vs Opus 5.5's 10.1%), LVBench (89.0% vs 83.7%), SEC-bench Pro (90.5%), and WebExploitBench v1.2 (51.3% vs 37.1%).
But the table also shows Opus has not lost outright
According to the very table in the post's image, "beats on most fronts" is not the whole story: Opus 5.5 still wins Terminal-bench 4.1 (65.8% vs Neon's 65.2%) and FrontierCode 2.0 (47.3% vs 46.2%), while GPT-6.2 Sol wins Chartography (75.1% vs Neon's 73.5%). And on the Agentic Knowledge test (AA-Briefcase, Elo scale), Claude still leads: Opus 5.5 Max scores 1800, Sonnet 5.5 Max 1815, while Neon xhigh stops at 1700. The author also notes Claude's subsidy is about 3 times better at the same cost as Gemini, so using Gemini now still burns Usage faster than Claude — the author expects Google to improve this, or Claude to cut prices after its IPO.
Apply now: Opus plans, Hydrogen/Neon execute
According to the post, if you are not sure about switching entirely to Neon for everything, the combo that still works very well is: use Opus 5.5 for planning and direction (such as audits, topical maps, SEO strategy), then hand execution to Hydrogen (super cheap, up to 92% non-hallucination) or Neon (code, complex tasks).
Frequently asked questions
1. On which index does Gemini 4 Neon beat Opus 5.5? According to the post, Neon scores 61 (xhigh) on the Artificial Analysis Intelligence Index, ahead of Opus 5.5 Max (59); long-context GraphWalks reading reaches 95.2% vs 66.8%.
2. How much cheaper is Neon than Opus 5.5? According to the post, Neon (high) costs $3.74/task for 60 points, nearly 2.7 times cheaper than Opus 5.5 Max ($10.04 for 59 points); the medium tier at $1.90 for 55 points is over 5 times cheaper than Sonnet 5.5 Max at the same score.
3. Does Neon win every test? No. According to the table in the post's image itself, Opus 5.5 still wins Terminal-bench 4.1 and FrontierCode 2.0, GPT-6.2 Sol wins Chartography; and on the Agentic Knowledge test (AA-Briefcase), Claude still leads (Opus 5.5 Max 1800, Sonnet 5.5 Max 1815, Neon xhigh 1700).
4. What are the non-hallucination rates? According to the post, the Gemini 4 line takes the whole top (Hydrogen 91%–92%, Neon 81%–86%), while Opus 5.5 manages only 33%–50%.
5. Which combo does the author recommend? According to the post: Opus 5.5 for planning and direction (audits, topical maps, SEO strategy), then Hydrogen (super cheap, up to 92% non-hallucination) or Neon (code, complex tasks) for execution.
Note on model names: the post writes "Claude Code + Fable 5.2" scoring 58 on the Coding Agent Index; Wotech could not confirm any model named "Fable 5.2", so this is quoted exactly as the post names it, without asserting it is a real model name. Source: post by Đỗ Anh Việt in the Nghiện Claude group (19/11/2026), figures per Google and Artificial Analysis; the benchmark table in the image cites its methodology from Google's eval page. All figures, views, and recommendations in this article follow the post and have not been independently verified by Wotech.
Bài viết liên quan

20 tháng 11, 2026 · 10 phut doc
Java Dev, 32, Laid Off 23 Days Before His Wedding: 187 Applications, 0 Offers, and a 404 Not Found
A 32-year-old Java developer in Hangzhou was laid off exactly 23 days before his wedding — and the ending is what leaves you sitting in silence. The story was shared by the account Stanley (@Stanleys…

20 tháng 11, 2026 · 5 phut doc
OpenClaw Creator Barely Lets Claude Write Code: Claude Plans, Codex Executes
Peter Steinberger, the creator of OpenClaw, almost never uses Claude to write code directly. Instead, he has a more efficient workflow: Claude plans, Codex executes. That is the content of a post by …

19 tháng 11, 2026 · 7 phut doc
First Director Criminally Tried in Vietnam for Using Unlicensed Software in His Own Company
No selling pirated software, no distributing cracked software — only using software to run his own company's operations. So why was the director still brought to criminal trial? On 17/11/2026, the Pe…