Gemini 4 Argon Early Access: 1M-Token Single Responses Spark Wave of 3D Demos
Google DeepMind opened early access to Gemini 4 Argon to a group of creators, and on October 14 several developers and bloggers published hands-on results. The core highlight is output capacity of up to 1 million tokens in a single response, along with strong performance in code generation and multimodal creation.
Confirmed
- The Prompt Engineering channel, granted early access by Google DeepMind, used the continuation token mechanism to break 1 million tokens in a single response, wrote out the entire contents of a seaside city in one 1-million-token response, and reproduced a paper to train an ML model.
- YouTube creator Sam Witteveen generated a ReBirth-style music production tool from one prompt, supporting 1-million-token output, also covering Blender and game generation, plus a "write code–render–view–fix" workflow.
- Developer fofrAI, while trying the code-drawing feature, casually asked for a photorealistic scene in pure Python; Argon wrote roughly 5000 lines of code, building a 3D ray tracing engine from first principles. The accompanying video, TTS voiceover, and background music (Lyria) were also produced by Argon.
- Developer @kickingkeys used Argon to generate a series of artworks — all 534 images drawn by rendered JS code rather than conventional image generation, with the code fully open-sourced; he considers Argon the first model that made him feel "image, video, and language model as one."
Why it matters
- Million-token single-response output dramatically expands the boundary of what models can accomplish in one go, making "deliver a complete app/scene from a single prompt" feasible.
- Multiple independent testers consistently report the model using code as a creative medium (drawing, music tools, 3D engines), signaling a trend of deep integration between multimodal and coding capabilities — worth watching after official release.
2026-10-14 ~ 2026-10-15 · 18 related posts
- Episode 1: Google Unveils Gemini 4 Argón with Focus on Deep Reasoning(2026-10-07, 2 posts)
- Episode 2: Gemini 4 Argon's Delayed Rollout Fuels Speculation of Weekend Release(2026-10-07, 4 posts)
- Episode 3: Gemini 4 Codenamed Argon Leaks Across Google Platforms Ahead of Launch(2026-10-09, 10 posts)
- Episode 4: 'Gemini 4 Argon' Hype Sweeps Reddit, But It's All a Joke(2026-10-10, 3 posts)
- Episode 5: Google reportedly tests Gemini 4 "Carbon" internally, coding nears Opus 5.5(2026-10-10, 19 posts)
- Episode 6: Gemini 4 "Argon" rumors surge as model option surfaces in Antigravity IDE(2026-10-11, 6 posts)
- Episode 7: Gemini 4 Argon Leaks Across Google Surfaces as Launch Nears(2026-10-12, 20 posts)
- Episode 8: Leaked Promo Images and UI Screenshots Suggest Imminent Gemini 4 "Argon" Launch(2026-10-13, 7 posts)
- Episode 9: Google launches Gemini 4 Argon across all products; robot learns cornhole from video(2026-10-13, 278 posts)
- Episode 10: Gemini 4 Argon Reviews Beat Expectations, but Paywalls and Quotas Spark Backlash(2026-10-14, 62 posts)
- Episode 11: Gemini 4 Argon Early Access: 1M-Token Single Responses Spark Wave of 3D Demos(2026-10-14, 18 posts)
- Episode 12: Gemini 4 Argon sweeps benchmarks: physical world, science, reasoning top spots(2026-10-14, 9 posts)
- Episode 13: Gemini 4 Argon Generates Playable 3D Games in One Prompt(2026-10-14, 2 posts)
- Episode 14: Gemini 4 Argon Arrives on GitHub Copilot(2026-10-14, 2 posts)
- Episode 15: Neel Nanda: Gemini 4 Argon markedly harder to monitor(2026-10-14, 2 posts)
- Episode 16: Gemini 4 Scores 3.95 in Blind Coding Audit, Developers Report Heavy Hallucinations(2026-10-15, 2 posts)
Primary sources
- Gemini 4 Argon Hands-On: 1M Output Tokens in One Response, Screenshot-to-Website and Autonomous ML — Prompt Engineering ·
- Gemini 4 Argon tested: one-prompt music maker, 1M token output and pricing breakdown — Sam Witteveen ·
- Gemini 4 Argon builds six playable 3D web games in one shot — all code open-sourced — ammaar ·
- [source] Gemini 4 Argon tested: one-prompt music maker, 1M token output and pricing breakdown — Sam Witteveen · 2026-10-14
- [source] Gemini 4 Argon Hands-On: 1M Output Tokens in One Response, Screenshot-to-Website and Autonomous ML — Prompt Engineering · 2026-10-14
- Dev renders 534 paintings entirely as JS code with Gemini 4 Argon, releases full source — SergioPaniego · 2026-10-14
- Gemini's Argon wrote 5,000 lines of Python to build its own ray-tracing engine — fofrAI · 2026-10-14
- Gemini 4 Argon wrote 5,000 lines of Python, building a 3D ray tracer from scratch — fofrAI · 2026-10-14
- [source] Gemini 4 Argon builds six playable 3D web games in one shot — all code open-sourced — ammaar · 2026-10-14
- Third entry in ammaar's Gemini game series: a VHS-horror corridor that jump-scared his team — ammaar · 2026-10-14
- All six Gemini-built games, prompts, and source now in one open GitHub repo — ammaar · 2026-10-14
- One prompt, 5,000 lines: Argon builds its own 3D ray-tracing engine in pure Python — technollama · 2026-10-14
- Gemini 4 Argon demos: 1M-token city build, paper-to-neural-net training, screenshot site rebuilds — Prompt Engineering · 2026-10-14
- Google ships Gemini 4 Argon; users already build 3D worlds and games with it — FinanceYF5 · 2026-10-15
- One image becomes a gothic 3D world with dynamic lighting via Argon — FinanceYF5 · 2026-10-15
- Argon writes its own 3D raytracer, video tool and audio mixer — FinanceYF5 · 2026-10-15
- Gemini 4 Argon users are building full 3D worlds and games: 10 wild examples — aitrendz_xyz · 2026-10-15
- Gemini 4 Argon spawns flood of 3D demos one day after launch: 10 examples — socialwithaayan · 2026-10-15
- One prompt gets Gemini 4 Argon to build a walkable 3D city in iPhone POV — socialwithaayan · 2026-10-15
- Purported Gemini 4 builds a shakable New York snow globe entirely from code, zero assets — prasenx · 2026-10-15
1 near-duplicate retellings: prasenx