Local AI HTML Coding Challenge
The Challenge (128GB)
The aim of the coding challenge is to provide an overview which freely available model performs best for the task on PC with 128GB shared memory. Focus is on the look & feel of the solution and the speed of the creation process.
Prompt: Create a stunning audio visualization for MP3 files in HTML. Allow to select the local MP3 file in the beginning.
Test environment: HP Z2 Mini G1a with AMD Ryzen Max+ PRO 395 and 128GB shared RAM running LM Studio 0.4.6 on Windows 11.
Summary
| Model | Quant | Size | Result |
| Qwen/Qwen3-coder-next | Q4_K_M | 48.5 GB | 👍 👍 🏎 |
| MiniMax-M2.5 | Q3_K_S | 98.7 GB | 👍 👍 🚗 |
| MiniMax-M2.1 | Q3_K_S | 98.7 GB | 👍 👍 🚗 |
| NVIDIA/nemotron-3-super | Q4_K_M | 86.1 GB | 👍 👍 🦆 |
| OpenAI/gpt-oss-120B | MXFP4 | 63.4 GB | 👍 🏎 |
| zai-org/GLM-4.7-flash | Q4_K_M | 18.1 GB | 👍 🦆 |
Look & Feel
Metrics
Qwen/Qwen3-coder-next Q4_K_M
ElapsedSeconds: 114.923 🏎
PromptTokens: 32
CompletionTokens: 3832
TotalTokens: 3864
TokensPerSecond: 33.623
MiniMax-M2.5 Q3_K_S
ElapsedSeconds: 240.194 🚗
PromptTokens: 63
CompletionTokens: 3455
TotalTokens: 3518
TokensPerSecond: 14.646
MiniMax-M2.1 Q3_K_S
ElapsedSeconds: 234.218 🚗
PromptTokens: 63
CompletionTokens: 3358
TotalTokens: 3421
TokensPerSecond: 14.606
NVIDIA/nemotron-3-super Q4_K_M
ElapsedSeconds: 281.998 🦆
PromptTokens: 40
CompletionTokens: 2426
TotalTokens: 2466
TokensPerSecond: 8.745
OpenAI/gpt-oss-120B MXFP4
ElapsedSeconds: 112.68 🏎
PromptTokens: 91
CompletionTokens: 2459
TotalTokens: 2550
TokensPerSecond: 22.63
zai-org/GLM-4.7-flash Q4_K_M
ElapsedSeconds: 278.342 🦆
PromptTokens: 29
CompletionTokens: 6207
TotalTokens: 6236
TokensPerSecond: 22.404
Variance
The prompt was run three times per model tom emsure consistence results. The time erquired to produce the solution matches plus/minus 5 seconds the values listed here. The HTML code produced was identical for every run and every model with LM Studios default settings.
