谷歌又杀回来了！Gemini 4 Argon 正式发布，实测有多强？

Summary
In this review video, tech creator 零度解说 (Lingdu Jieshuo) introduces and tests Google’s newly released flagship AI model, Gemini 4 Argon. He analyzes official benchmarks and showcase demos—including robotics control, 3D procedural generation, and codebase maintenance—before running hands-on tests in the Gemini web interface and demonstrating how to renew the Google AI Pro student plan.

---

What is shown

• [00:00–00:58] Introduction of Gemini 4 Argon, highlighting benchmark leaderboards (LiveBench 2.0, Vals Index for finance/law/tax).
• [00:59–01:37] Robotic demonstration of Gemini 4 Argon controlling an Apptronik Apollo humanoid robot to aim and throw beanbags into a cornhole board in real-time.
• [01:38–02:06] Codebase maintenance capabilities tested against the Chromium open-source repository.
• [02:07–03:08] Code-driven procedural 3D model generation of a sci-fi attack helicopter (STRIX HX-9) with animations, lighting, and flight simulation without standard 3D modeling software.
• [03:09–03:36] Cost and speed comparison demo for a 3D space train generation prompt between Gemini 4 Argon and Claude Opus 5.5.
• [03:44–05:18] Practical Test 1: Feeding 4 reference photos into Gemini 4 Argon to generate an interactive Three.js 3D virtual photography studio with customizable backdrops, camera presets, and dynamic lighting.
• [05:19–06:32] Practical Test 2: Generating a complete interactive 10-year “Rent vs. Buy” financial simulator web dashboard with dual-bound sliders, amortization schedules, and CSV export.
• [06:33–08:02] Multi-agent Blender orchestration demos, including procedural Chinese temple architecture, a 174,000-polygon Gothic Victorian carriage, and “Desk Racer” (converting a photo of a desk into a playable Three.js racing game).
• [08:15–08:31] Image generation test in the Gemini interface producing a photorealistic portrait.
• [08:41–09:16] Downloading and setting up the Google Antigravity desktop client on Windows using TDM Fast.
• [09:21–18:09] Step-by-step tutorial on renewing/re-verifying the free 1-year Google AI Pro student plan (5TB Google One storage, Gemini access) using Proton VPN (US endpoint) and a Binance virtual payment card, including how to cancel subscription auto-renewal while retaining benefits through August 2027.
• [18:10–18:31] Testing video generation inside Gemini (generating a video clip of a young couple studying in a library with synchronized English dialogue).

---

Claims & numbers

• The presenter notes Gemini 4 Argon officially launched on the evening of October 13, 2026.
• In LiveBench 2.0, the presenter states Gemini 4 Argon scored an overall 84.1 (5th place), while ranking 1st in Reasoning with 95.1 (leading 2nd-place GPT-6 Astra at 92.7 by 2.4 points) and 1st in Data Analysis with 85.2.
• On the Vals Index (financial, legal, tax workflows), Gemini 4 Argon scored 68.9%, ranking 1st.
• In the 3D space train generation benchmark, the presenter claims Gemini 4 Argon took 4 minutes and 20 seconds at a cost of $3.40, while Claude Opus 5.5 took 1 hour, 52 minutes, and 56 seconds at a cost of $23.46 (making Gemini Argon ~26x faster and nearly 7x cheaper).
• The procedural Gothic Victorian carriage generated entirely via code features 174,000 polygons.
• The presenter states the free Google AI Pro student plan provides 12 months free (normally $19.99/month), 5TB of Google One cloud storage, and quadrupled usage limits compared to the free tier.

---

Notable quotes

• [00:29] “第1个是逻辑推理，拿下了95.1的高分，全场排名第1。” (“First is logical reasoning, capturing a top score of 95.1 to rank first across the board.”)
• [03:21] “速度快了26倍，而且成本降低了将近7倍。” (“The speed is 26 times faster, and the cost was reduced by nearly sevenfold.”)
• [08:03] “Gemini 4 Argon是第1个让我感觉像是图像、视频和语言模型融为一体的AI大模型。” (“Gemini 4 Argon is the first AI foundation model that feels truly unified across images, video, and language.”)

---

Assessment
This is a third-party hands-on review and tutorial video combining commentary on Google’s launch demos with real-time web UI testing. The presenter successfully validates the model’s single-prompt web coding generation (the 3D photo studio and financial calculator) and subscription re-verification workflow live on screen, while relying on pre-recorded third-party and vendor clips for the complex robotics and Blender MCP demonstrations.

Described by gemini-3.8-flash on 2026-10-14 from the video’s audio and frames.