Token Gremlin
@TokenGremlin • 14,915 subscribers
Tracking the AI frontier: models, agents, research, leaks & pre-release signals. I test, investigate and build.
Shorts
Videos
Tibo confirmed that OpenAI plans to merge Codex and ChatGPT capabilities, with the goal of making the experience less complex for users: “…and for Codex and ChatGPT, we have Chat and the Work toggle, and we’re going to merge them. We’ve heard the feedback loud and clear: people love the capabilities in Work, but sometimes you want to use Chat because it’s a little faster and more pleasant to use. So we’re merging the two and reducing that complexity.”
Token Gremlin223,597 次观看 • 24 天前
I’ve officially open-sourced ABYSSAL. 🌊⚡ ABYSSAL is a fully procedural ocean and extreme-weather experience that runs directly in the browser. Waves, volumetric clouds, rain, lightning, hurricanes, waterspouts, rogue waves, and tsunamis are generated in real time using Three.js, WebGL2, and GLSL. The project is still actively being developed, and there’s a lot left to do. Deeper optimization, building the world beneath the ocean’s surface, refining the existing systems, and exploring plenty of ideas I haven’t implemented yet. But that’s also why I wanted to open it now. ABYSSAL is MIT-licensed, so you can explore the code, learn from it, modify it, fork it, and use it as the foundation for your own experiments and projects. Performance note: the demo starts with adaptive resolution enabled and automatically selects a quality tier based on your GPU. If the image looks soft, open the settings panel and raise the quality as far as your device can handle. The experience begins in Cinematic mode, click SANDBOX for full control of the camera, weather, and disasters. Explore the source code: Play it here: Feedback, contributions, experiments, and forks are all welcome.
Token Gremlin257,063 次观看 • 2 个月前
A lot of people have been telling me they’re seeing noticeable improvements in both quality and latency with GPT-6 Astra and GPT-5.6 Sol. I’m still looking into it, but it really does seem like something changed. Both models suddenly feel much better “out of nowhere.” I’ve also started seeing reports and examples of much more advanced interactive interfaces being generated directly inside ChatGPT, like in the video below. I still don’t know if these changes are connected, but something is clearly moving behind the scenes.
Token Gremlin21,926 次观看 • 1 个月前
GPT-6.1 Sol, running in Ultra mode, has been autonomously iterating on this project for more than 16 hours now. It’s using Blender and Three.js, and the entire thing runs directly in the browser. Like I said earlier, this model is genuinely impressive. OpenAI built something here that feels extremely capable, surprisingly cheap, and now fast enough to make long-running agentic workflows much more practical. (The project is still in progress...)
Token Gremlin18,553 次观看 • 28 天前
A really impressive set of Three.js graphics experiments just got open sourced, and these are much more than little visual demos. They are basically reusable procedural systems for oceans, vegetation, fluids and even whole planets. 🔹 Poseidon A real-time FFT ocean running on WebGPU. It simulates large swells, smaller ripples, foam, reflections, choppy displacement and physically inspired wave spectra entirely in the browser. 🔹 Gaia A procedural grass generator where every blade, seed head and field comes from a deterministic genome and environmental parameters. No authored grass models. 🔹 Dryad A procedural flora system that generates trees and other plant forms from physics, environmental conditions and a seed. No authored 3D models or textures are needed for the plants themselves. 🔹 Tiamat A real-time GPU fluid simulation using around 100,000 SPH particles, with the resulting water rendered directly in the browser. 🔹 Demiurge Probably the craziest one. It procedurally builds an entire planet from tectonic plates, then lets uplift drive erosion, erosion and latitude drive climate, and climate drive biomes, wind and weather. You can move seamlessly from orbit down to the surface. What I really like here is that these are not just pretty outputs. They are actual building blocks. Ocean simulation, vegetation generation, fluid dynamics and procedural worlds are exactly the kinds of systems that can be plugged into games, simulations and agent-built 3D environments. Project by: Owen
Token Gremlin34,448 次观看 • 2 个月前
GPT-5.6 Sol Ultra was given a full set of character references and asked to rebuild the character in Blender, rig it, create three animations, export a real GLB, and build a complete Three.js showcase around it. The entire pipeline was completed in 2 hours and 11 minutes. The final model still loses some of the original proportions, costume structure, and material fidelity, but going from reference images to a rigged, animated 3D character running in the browser is already a very strong result. Cc: Givros
Token Gremlin37,583 次观看 • 3 个月前
I built an entire Gothic basilica with Opus 5.5, and I’ve open-sourced it. ⛪✨ Basilica of Celestial Light is especially close to my heart. I genuinely loved creating this church. Working on it brought me a real sense of peace, and being able to share it with everyone feels just as meaningful. Everything was created with Opus 5.5, entirely in code. No external assets: no imported 3D models, image textures, or audio files. The architecture, stained glass, landscape, lighting, and soundscape are all generated at runtime with Three.js. You can walk through the nave, visit the crypt, climb the bell tower, explore the gardens, or take the guided tour, all directly in your browser. The project is open source under the MIT license. Explore it, learn from it, change it, fork it, and make something of your own. I’d genuinely love to see what others create from it. Source code: Play it here: I hope you find a little of the peace I felt while creating it.
Token Gremlin13,921 次观看 • 1 个月前
I’ve officially open-sourced SYLVA. 🌲 SYLVA is a fully procedural cinematic forest that runs directly in the browser. Terrain, trees, roots, grass, water, sky, clouds, fog, and weather are generated at runtime using Three.js, WebGL2, and GLSL, without downloaded tree models, textures, HDRIs, or terrain heightmaps. Trees are procedurally grown, terrain is generated mathematically, and the world includes dynamic water, atmosphere, vegetation LODs, rain, lightning, wind, storms, fireflies, birds, and a cinematic camera system. The demo is also interactive. Open the EDITOR to change the environment, visual presets, quality, weather states, wind and storms, move through different times of day, or simply walk around and explore the forest yourself. The project is still actively being developed, and there’s a ridiculous amount left to improve, optimize, and experiment with. But that’s also why I wanted to open it now. SYLVA is MIT-licensed, so you can explore the code, learn from it, modify it, fork it, optimize it, or use it as the foundation for something completely different. Source code: Live demo: Feedback, contributions, experiments, and forks are all welcome. 🌲 (This project is being updated RIGHT NOW.)
Token Gremlin14,608 次观看 • 1 个月前
Fable 5 built this real-time ocean scene directly in the browser using Three.js, WebGPU and TSL, with no engine and no baked assets. The interesting part is not just that the water looks good. The whole system is being simulated: wave spectrum, GPU FFT, foam generation and refracted-ray caustics, all running live. This is exactly where frontier models start becoming genuinely useful for technical graphics work. Not by replacing years of rendering research, but by compressing the path from complex math to a working interactive result. A small, focused scene, but technically very serious. Cc: Adem Vessell
Token Gremlin21,731 次观看 • 3 个月前
Same prompt. GPT-5.6 Sol vs Opus 5. All in Three.js. Opus produced the better raw visuals. The ocean, waves and atmosphere look noticeably richer. But it also: → cost ~10x more → took much longer → needed an extra prompt just to fix the camera controls Sol got surprisingly close on the first attempt, much faster and far cheaper. Opus wins visual fidelity. Sol wins the actual engineering test.
Token Gremlin12,914 次观看 • 2 个月前
Claude Opus 5 built a Red Dead Redemption-style open world western in Three.js with just a few extra prompts added to a gauntlet loop. The result is a surprisingly strong environment with horseback traversal, large-scale terrain, atmospheric lighting, and a level of polish that already looks better than a lot of small indie prototypes. Play it: Repository: Project by: Michael Gill
Token Gremlin15,182 次观看 • 2 个月前
Kimi K3 built a browser-based Windows XP environment, and it goes way beyond a visual mockup. It recreates the desktop, system UI, and old software behavior closely enough to feel like a usable web-native OS layer. It can even run games like GTA: Vice City and Red Alert 2. That is a much more interesting result than a simple nostalgic interface demo. Cc: AB Kuai.Dong
Token Gremlin16,113 次观看 • 3 个月前
One of my favorite things about open source is seeing someone take an idea and run with it. The original ABYSSAL started as a procedural ocean and extreme-weather simulation. Now someone has built "ABYSSAL - The Living Deep" on top of it: a continuous world from the waterline to a 1,400-meter trench, with coral reefs, kelp forests, whales, sharks, deep-sea creatures, procedural habitats, currents, lighting and exploration. This is why open source is worth it. People can contribute from different directions, build on the same work, and reach results that would be difficult to imagine alone, sometimes in an incredibly short time. The original ocean is still there. It just became something much bigger. Explore it here: Live demo:
Token Gremlin10,288 次观看 • 1 个月前
At this point, Three.js is probably the best graphics stack for AI-generated games. Not because it magically produces better visuals, but because models are extremely good at reasoning directly over JavaScript, scene graphs, shaders, procedural geometry, physics integrations, and fast browser iteration. The examples here make the case pretty clearly. Once multiple agents start building scenes in isolation and integrating them later, the ceiling gets surprisingly high. For AI-native game development, Three.js currently feels like the path of least resistance to genuinely impressive results. I gathered a few videos from Chetan in this demo and combined them.
Token Gremlin12,038 次观看 • 2 个月前
Fable 5 built this 3D world in Three.js with no external assets. The project was almost entirely generated autonomously by Fable 5, with only occasional human feedback for visual details the model could not reliably inspect, like moving water and wind effects. That makes the result a lot more interesting than a normal assisted 3D demo: the terrain, vegetation, atmosphere and world structure were effectively produced by the model itself. The full prompt is also included in the repository, so the experiment is inspectable. Repository: Project by: Remi Sebastian Kits
Token Gremlin11,708 次观看 • 2 个月前
没有更多内容可加载