Pipeline OS

Exploring the future of AI & tech. ☁️
🧠 Testing models, tracking updates (and calling out the nerfs). Welcome to the Pipeline. 🚀

[Pipeline OS @pipeline_os]

🚨Gemini 4 Pro vs. Fable 5.1 (in Arena)

Gemini destroys Fable by a massive margin.

Waiting for the official release — with performance like this, it will be an absolute game over for the market.

[Pipeline OS @pipeline_os]

🚨New tier just dropped, and only one model is in it.

Gemini 4 Argon didn't just top the list, it needed its own row above SSS 😭

Which model is ranked too low?

[Pipeline OS @pipeline_os]

Animators spend weeks on a running cycle like this.

Three AI models just did it for the same hippo.

> Gemini 4 Argon gave it a face with personality
> Claude Opus 5.5 made the water splash feel alive
> GPT-6 Sol went for a calm, minimal sunset

Which one would you keep?

[Pipeline OS @pipeline_os]

Animators spend weeks on a running cycle like this.

Three AI models just did it for the same hippo.

> Gemini 4 Argon gave it a face with personality
> Claude Opus 5.5 made the water splash feel alive
> GPT-6 Sol went for a calm, minimal sunset

Which one would you keep?

[Pipeline OS @pipeline_os]

🚨Gemini 4 Argon is #5 in Agent Arena.

Then I asked it to build a beach house you can walk through.

> Two floors and a beach lounge
> Sunset light through the glass
> Time of day you can switch
> Everything runs in a browser

Leaderboards don't tell the whole story.

What should I make it build next?

[Pipeline OS @pipeline_os]

🚨Google's own table says Gemini 4 Argon leads 13 of 19 benchmarks.

Arena's Agent leaderboard puts it 5th.

> 1. Claude Opus 5.5
> 2. GPT-6.1 Sol
> 3. Claude Fable 5.1
> 4. GPT-6 Astra
> 5. Gemini 4 Argon

Both are real data. They just measure different things.

Which one do you trust more?

[Pipeline OS @pipeline_os]

🚨Google's own table says Gemini 4 Argon leads 13 of 19 benchmarks.

Arena's Agent leaderboard puts it 5th.

> 1. Claude Opus 5.5
> 2. GPT-6.1 Sol
> 3. Claude Fable 5.1
> 4. GPT-6 Astra
> 5. Gemini 4 Argon

Both are real data. They just measure different things.

Which one do you trust more?

[Pipeline OS @pipeline_os]

Create a complete, single-file HTML5 3D web application using Three.js (via CDN) that renders a high-quality, smooth 60fps voxel art animation.

STRICT REQUIREMENT: Do NOT include any HTML/CSS UI overlays, text, titles, buttons, or HUD elements. Fullscreen canvas only.

Scene details:
1. Voxel Fighter Jet: A detailed grey voxel fighter jet with twin engines, cockpit canopy, wings with red/blue markings, and missiles.
2. Voxel Pelican on a Bicycle: On top of the jet, build a white voxel pelican wearing black sunglasses and a waving red scarf, riding a red-and-yellow voxel bicycle posed in a wheelie. Animated pedaling legs and spinning wheels.
3. Particle System: Dynamic voxel cube exhaust particles shooting from the jet's twin engines (glowing orange/yellow/red fire particles that fade and shrink).
4. Environment: Gradient skybox, directional sunlight with soft shadows (PCFSoftShadowMap), ambient lighting, and procedural white boxy voxel clouds with visible grid outlines scrolling past rapidly to create a sense of high speed.
5. Animation & Controls: Smooth idle flight motion (gentle bobbing and banking), periodic automatic barrel rolls (or triggered via Spacebar), and Three.js OrbitControls with damping enabled for smooth camera rotation and zooming.
6. Graphics Quality: Use InstancedMesh or merged geometries for high performance, high-DPI renderer support (window.devicePixelRatio), antialiasing enabled, and crisp voxel shading.

[Pipeline OS @pipeline_os]

🚨Woow, Gemini 4 Argon really came out better than its competitors.

Same scene: a rider flying through floating sky islands.

> Gemini 4 Argon
> Claude Haiku 5.5

Argon has the richer sky, the jet flames and more detailed islands. Haiku keeps it simple.

Who should I compare it with next?

> Claude Opus 5.5
> GPT-6 Astra
> Claude Fable 5.1

[Pipeline OS @pipeline_os]

@_mateusz_el @GeminiApp @Google @AnthropicAI What do you want to see next?

[Pipeline OS @pipeline_os]

🚨Woow, Gemini 4 Argon really came out better than its competitors.

Same scene: a rider flying through floating sky islands.

> Gemini 4 Argon
> Claude Haiku 5.5

Argon has the richer sky, the jet flames and more detailed islands. Haiku keeps it simple.

Who should I compare it with next?

> Claude Opus 5.5
> GPT-6 Astra
> Claude Fable 5.1

[Pipeline OS @pipeline_os]

Google says Gemini 4 Argon cut code rollbacks by nearly a third and lifted agentic code submissions by 35% in Antigravity.

Those are Google's own numbers, on Google's own codebases.

One early outside tester says Antigravity still falls short of Claude Code and Codex.

Both can be true. The next few days of real use will tell.

Who's switching?

[Pipeline OS @pipeline_os]

In the 146 days since May 19, this chart counts 18 model releases from three labs.

> Anthropic: 8
> OpenAI: 7
> xAI: 3

That's about one every 8 days.

Google announced Argon on Sep 30 and access is still limited.

Is Google playing a different game, or just late?

[Pipeline OS @pipeline_os]

🚨Claude Opus 5.5 built an apartment you can walk inside.

> Walk every room
> Open the front door
> Morning → afternoon → sunset

Now let's see what it does with Blender.

Want to see Claude + Blender?

[Pipeline OS @pipeline_os]

🚨Claude Opus 5.5 built an apartment you can walk inside.

> Walk every room
> Open the front door
> Morning → afternoon → sunset

Now let's see what it does with Blender.

Want to see Claude + Blender?

[Pipeline OS @pipeline_os]

🚨Gemini 4 and Gemini 4 Flash could be next in line.

Google seems to be building the whole lineup around Argon:

> Argon was announced on Sep 30, with limited access so far
> Antigravity already shows Argon upgrade prompts
> Reports say Gemini 4 and Flash may follow, but nothing is confirmed

12 days after the announcement, the pieces look ready. The date is still missing.

Will Gemini 4 and Flash launch with Argon, or months later?

[Pipeline OS @pipeline_os]

🚨Game studios spend months building an ocean you can swim in.

I gave Claude Opus 5.5 one prompt.

> A fish you can control
> A camera you can orbit
> Light that moves across the seabed

Which part surprised you most?

[Pipeline OS @pipeline_os]

@FReza1984 Fair point. Even on paper the coding rows are mixed: Argon leads DeepSWE and Vibe Code Bench, but loses FrontierSWE v2 and Terminal-bench 4.0. Messy real tasks will decide it.

[Pipeline OS @pipeline_os]

🚨Google's own table shows Gemini 4 Argon ahead of GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 on most rows.

Google's own table.

> Leads 13 of 19 benchmarks
> Ties 1
> Loses 5, two of them to Claude Opus 5.5

Now let's see what it does on real tasks.

[Pipeline OS @pipeline_os]

🚨Google's own table shows Gemini 4 Argon ahead of GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 on most rows.

Google's own table.

> Leads 13 of 19 benchmarks
> Ties 1
> Loses 5, two of them to Claude Opus 5.5

Now let's see what it does on real tasks.

[Pipeline OS @pipeline_os]

@ajs6888 Fair point. Each line goes from the cheapest plan to the most expensive:

Anthropic $20 / $100 / $200 = S / A / S
OpenAI $8 / $20 / $200 = D / A / B
Google $7.99 / $19.99 / $249.99 = C / C / F

[Pipeline OS @pipeline_os]

🚨 Don't buy any AI subscription until you see this tier list.

Anthropic: S, A, S
OpenAI: D, A, B
Google: C, C, F

Gemini Ultra at $249.99 got the only F among the AI plans.

Do you agree?