Gemini 4 Argon destroyed me. I’m done.
GothamChess YouTube 747,077 views as of 17 October 2026
Why it is here
Chess YouTuber Levy Rozman (GothamChess, 27 min) on Gemini 4 Argon, following his earlier GPT-6 Astra and Opus 5.5 chess videos in this dataset. The description gives no details of the games, so what he tested is pending the video description. About 747k views in a day.
Description
Description written by Gemini from the videoGemini 3.8 Flash, 17 October 2026
Summary
Chess International Master and creator Levy Rozman (GothamChess) reviews two chess games he played against Google DeepMind’s Gemini 4 Argon. He analyzes the model’s tactical strengths, aggressive play, and sudden hallucinations—including illegal moves and an endgame blunder where the bot sacrificed its own king.
What is shown
- [00:00] Levy Rozman introduces the video, discussing the recent evolution of LLM chess capabilities and introducing his match against Gemini 4 Argon.
- [01:04] Chess.com board analysis interface showing Game 1: Levy (White, 2320) vs. Gemini Argon 4 (Black).
- [01:42] Demonstration of Game 1 opening (Trompowsky Attack), where Levy gambits pawns and traps Gemini’s queen on the kingside.
- [09:22] Gemini finds a queen sacrifice (Qxh5) for a rook and knight, entering an endgame where engine evaluation drops to dead equal or slightly favorable for Black.
- [11:35] Gemini collapses tactically, attempts illegal moves, and gets checkmated at [11:52].
- [12:34] Game 2 analysis: Gemini plays White against Levy’s Caro-Kann Defense, executing queenside castling and an aggressive pawn storm on the kingside (h4–h5).
- [18:15] Tactical breakdown showing where Gemini missed winning sacrifices (Rxe6), allowing Levy to trade queens at [20:52].
- [25:01] Game 2 conclusion where Gemini hallucinates an illegal king move into check (Kd2/Ke2), followed by Levy hitting a 3-hour rate limit lockout.
- [26:30] Levy recaps his impressions and announces upcoming tour stops in Europe starting October 25.
Claims & numbers
- Levy states that LLM chess capabilities leaped significantly in mid-October 2026 with models such as GPT-6 Astra, Claude Fable, Claude Opus, and Gemini 4 Argon [00:23].
- In Game 1, Stockfish evaluated Gemini’s fortress position following its queen sacrifice as equal despite being down material [09:44].
- In Game 2, Gemini reached a winning evaluation of over +3 against Levy’s Caro-Kann before blundering by trading queens [15:28].
- Gemini locked Levy out of the interface with a 3-hour cooldown limit for exceeding usage [25:27].
- Levy claims he has met 500 to 600 people over the past two weeks across book tour events and the Grandmasters premiere [26:36].
Notable quotes
- [00:23] “But this week, here in mid-October 2026, I was shocked at the capabilities of Astra by GPT, and then also by Claude’s Fable and Opus...”
- [05:03] “It’s weird to play a chess game where you feel like no matter what you do, you are going to win, right? And this is how I feel when I play Gemini.”
- [25:16] “So it decided that the best way to end this game was not to lose its rook, but to sacrifice its king to me and lock me out for three hours because I ran out of usage.”
Assessment
This is a genuine gameplay review and analysis video demonstrating authentic interactive sessions between an International Master and Gemini 4 Argon. The board analysis reproduces actual moves made by the model, capturing both its high-level positional play and typical LLM failure modes like rule hallucinations and illegal king moves.
Described by gemini-3.8-flash on 2026-10-17 from the video’s audio and frames.
Related
More videos on this
-
Start Building with Gemini 4 Argon
Google Gemini10,659,493 views as of 17 Oct
-
Turn Your Webcam Into a Playable Synthesizer with Gemini 4 Argon
Google Gemini5,891,712 views as of 17 Oct
-
Gemini 4 Argon: 3D World in Blender
Google Antigravity4,168,049 views as of 17 Oct
-
Gemini 4 Argon is now globally available
Google DeepMind (@GoogleDeepMind)launch film2,500,961 views as of 17 Oct