Browser-Managed
Prompt API
Browser-selected model
- Limited browser support
- May be too small for the task

AI Poland - 19 November 2026
Rainer Hahnekamp
Multikino Złote Tarasy - Warsaw, Poland





Why local?
No provider charge per token.
Keep code and context inside.
Reduce provider dependency.

01



Browser-selected model
General ML
LLM Inference

02
Google ADK
vLLM
Qwen
Muse Glimmer


Spring Boot + Angular · Measured on ASUS GX10
| Model | Weights | Quantization | Time (h:mm) | Eff. Tokens/s | Validation | Review |
|---|---|---|---|---|---|---|
| Nemotron 3.5 Bolt 74B | 22.0 GB | IQ2 | 2:39 | 29.8 | 3/5 | 1/52 incomplete |
| Ternary Bonsai 2 27B | 5.9 GB | PTQ1_0 | 4:06 | 16.4* | 4/4 | 4/4 |
| GLM-5.3 Flash | 97.6 GB | IQ1_M | 1:28 | 13.4 | 5/5 | 5/5 |
| Muse Glimmer 30B | 15.9 GB | Q4_K_XL | 3:00 | 12.1 | 2/5 | 2/5 |
| Qwen3.8 27B | 17.6 GB | Q4_K_XL | 2:58 | 8.6 | 5/5 | 5/5 |
| Gemma 4 26B-A4B QAT | 14.2 GB | Q4_K_XL | 0:46 | 52.5 | 2/4 | 1/41 incomplete |
| Gemma 4 31B QAT | 17.3 GB | Q4_K_XL | 1:44 | 9.6 | 1/3 | 1/3 |
| DeepSeek V4 Flash | 86.9 GB | IQ1_M | 1:59 | 15.0 | 3/4 | 3/4 |
| Qwen3.6 27B | 17.6 GB | Q4_K_XL | 3:05 | 9.3 | 5/5 | 5/5 |
| Devstral Small 2 24B | 14.5 GB | Q4_K_XL | 5:23 | 9.3* | 1/2 | 1/2 |
| Qwen3-Coder 30B-A3B | 17.7 GB | Q4_K_XL | 0:34 | 36.9 | 0/5 | 0/5 |
47 runs · Medians · * Partial timing
18
GX10 Measurement × Memory Bandwidth ÷ 273 GB/s
| Device | Memory | Bandwidth | Eff. Tokens/s |
|---|---|---|---|
| ASUS GX10 | 128 GB | 273 GB/s | — |
| NVIDIA DGX Spark | 128 GB | 273 GB/s | — |
| MacBook Pro M5 Max | 128 GB | 614 GB/s | — |
| Mac Studio M5 Ultra | 256 GB | 1,200 GB/s | — |
| GeForce RTX 5090 | 32 GB | 1,792 GB/s | — |
| RTX PRO 6000 Blackwell | 96 GB | 1,792 GB/s | — |
Blue = measured on GX10 · ≈ = bandwidth estimate
Estimates assume model weights, KV cache, and runtime buffers all fit in available GPU memory.
BF16
K-Quant / UD-Q4_K_XL
Ideal 4-bit: 12.6 GB
128K / FP16 KV
Estimated KV cache
Model + context
Before runtime overhead






DE / AT Retail Examples · Incl. VAT · 19 November 2026
21