Self-playing Flappy Bird and Missile Command games, each one-shot in a single HTML file, comparing 4 open models
WebsitesDesignRead the post
empiriolabs.ai · web source · link re-checked 2026-10-16
Qwen 3.8 Max 0902
By Alibaba. 3 results on 3 boards, copied as published. Different boards, different metrics: don't add them up.
License (as listed)Proprietary
Hosting regionNot confirmed
Boards3
Code
No result on the boards we captured
Agentic / terminal
- 26.97%Terminal-Bench 4.0
Qwen3.8-Max-0902 (max) · Resolution rate
Claude Code · as of 2026-09-02 ↗
Websites
Design
No result on the boards we captured
Android
No result on the boards we captured
iOS
No mature public leaderboard
Available in
No plan row in our pricing data names this model version.
Use cases with this model
Same family: Qwen 10
Llama: only local infra (e.g., Llama 3.2 3B at 124 tok/s on a jailbroken PS5)
CodeRead the post
x.com · link re-checked 2026-10-16
"qwen 3.8 flash next … built the whole thing on my dgx spark in 2.5 hours", where Mistral Large 4 "burned 233k tokens and never wrote a single file"
CodeRead the post
x.com · link re-checked 2026-10-16
Data-analyst agent: "Qwen handles the main agent work", Vercel runs the loop
CodeRead the post
x.com · link re-checked 2026-10-16
Test: "Qwen 3.8 Max (max): 25.7 - DeepSeek V4.1 Flash (max): 21.7 - Kimi K3 (default): 23 - GLM-5.3 (max): 19 - Mistral Large 4 (high, its max): 15"
CodeRead the post
x.com · link re-checked 2026-10-16
So long Quickbooks! Thanks to local Qwen 3.8 27b (with some code review by Claude) I have built my own accounting system … $460 annual savings.
CodeRead the post
x.com · link re-checked 2026-10-16
A 5.9GB ternary compression of Qwen 3.8 27B building a frontend in one shot on an RTX 3060 12GB ("1 hour 24 minutes")
DesignRead the post
x.com · link re-checked 2026-10-16
Sources
lmarena.ai/leaderboard/webdev tbench.ai/leaderboard vals.ai/benchmarks/vibe-code