I used grok with Grok Build to continue developing my hobby game in threejs almost no errors
Very fast but being fast also makes me prompt more meaning spend more as well, it is both good and bad in a way :) Almost no errors as well, flawless…
This space is open for sponsorsThe slot is unsold. Support framewire on the donation page, or get in touch to book it. · FRAMEWIREHOUSELead dispatch
Alican KirazAug 1442
Advanced Frontier LLM Coding Benchmark 13.08.2026
Modeller: - Grok 4.6 Extra High - Cursor - Opus 5 Max - Claude Code (App) - Qwen 3.8 Max - Qwen-Code (CLI) - Kimi K3 Max - Kimi-Code (App) - GLM 5.2 Max - Zcode (App) - GPT 5.6 Sol Very High - Codex (App) -
Latest dispatches
qwen38Open any frame for the full clipNew harness (qwen 3.8) got screenshots of the account.
Told it to write a song. it did. minimax-music3 made the track. first output.
Gemini 3.7 Flash/Qwen 3.8 Max by just being told "make it as realistic as possible"
Just invented a novel test.
Grizzly bear riding a scooter in a coastal environment. Qwen 3.8 Max/Gemini 3.5 Flash Lite.
I gave qwen-3.8-max the ability to write music .
Here's the first song it wrote ☆*: .｡. o(≧▽≦)o .｡.:*☆
Alibaba tiene una IA que genera imágenes y vídeo gratis.
Sin tarjeta. Sin trial de 7 días. Gratis. ¡Y sin marca de agua! He visto la prueba completa de Qwen 3.8 Max y hay 3 cosas que no esperaba. Hilo 🧵
Alibaba dropped Qwen 3.8 last week, and it's now on DigitalOcean Serverless Inference.
The thing that got my attention is the context window, hundreds of thousands of tokens in one pass. Basically a whole codebase.
Japanese Tea Garden Bench - Qwen 3.8 Max
In the Text Arena, Gemini 3.7 Flash (High) landed #9 with 1490 pts, another improvement from Gemini 3.6 Flash (High) (#16 -> #9).
By category, Gemini 3.7 Flash (High) also strongly improved: - #3 Creative Writing (#12 -> #3) - #4 Math (#7 -> #4) - #9 Instruction Following (
Meta's Muse-Glimmer-30B is the first model running on my 3090 able to one shot this.
Not super complicated at all, all other models couldn't one-shot and needed more guidance to get to result.
The "I don't have enough VRAM" excuse just died.
I’m running Meta’s new 30B Muse Glimmer Q6_K_XL with a massive 130k context window on just 26GB VRAM FREE compute on Kaggle.
Qwen 3.8 just landed: 2.4 trillion parameters.
And a chat tab is the worst place to use it. The model is a monster: → Near the top of the largest models ever publicly released → Reads text,…
Grok 4.6 just joined the model lineup 👀
So we gave it the same traditional Chinese paper-cut style animation prompt alongside GPT-5.6 Sol, Claude Opus 5, and Qwen 3.8 Max.
Same prompt different output
Grok 4.6 - gpt 5.6 sol - qwen 3.8-max - Claude opus 5 who won ?
Qwen 3.6-27B just dropped.
There's now a free AI model that beats one 14 times its size. Alibaba's own 397 billion parameter flagship just lost to their new 27 billion one.
Qwen 3.8 Max Full COURSE 1 HOUR (Build & Automate Anything)
Grok 4.6 seems like a big improvement
I gave it, along with Claude Opus 5, GPT 5.6 Sol, and Qwen 3.8 Max, a banana.
Hermes agent OS just killed the “5 AI tabs” workflow.
The biggest upgrade isn’t another model. It’s what happens when every agent finally shares the same memory.
Grok 4.6 vs. DeepSeek V4 Pro vs.
Qwen 3.8 2.4T—a full-on game-building battle royale. DeepSeek was the only model to get the game right on its first attempt.