| Date | Model | Test result | Notes |
|---|---|---|---|
| July | Colibri GLM5.2 | in testing | |
| June | Kimi K2.7 CODE UD Q8-K-XL (602Gb Ram 32k context) | Success✅️ | good speed, makes working code 1st try Result |
| June | GLM 5.2 UD Q6-K-XL (~690Gb Ram 13k context) | Fail❌️ | very 2x slow vs kimik2,6 Result |
| June | Nemotron 3 Ultra 550B-A55B Q8-K-XL (600Gb Ram 16k context) | Fail❌️ | Very good in writing Novels, and very valuable hallucinations(*) content |
| June | Stepfun 3.7 Flash 196B-A11B Q8 (215Gb Ram 32k context) | Fail❌️ | very slow, super long in thinking, need special system prompt |
| June | Deepseek V4 Flash Q8 (315Gb Ram 65k context) | inconsistent🔀️ | it resolved the test and even writes music in ChucK but on restart failing |
| May | GLM 5.1 UD Q6-K-XL (730Gb Ram 65k context) | Fail❌️ | very different model by other tests compared to previous GLMs |
context size different and also consumes Ram, depends on total model size
ChucK code can be played online as sound on https://chuck.stanford.edu/ide/
| Date | Model | Test result | Notes |
|---|---|---|---|
| release date | Kimi K2.7 CODE UD Q8-K-XL (602Gb Ram 32k context) | Success✅️ | good speed, makes working code 1st try |
| release date | Deepseek V4 Flash Q8 (315Gb Ram 65k context) | inconsistent🔀️ | it resolved the test and even writes music in ChucK but on restart failing |
| release date | Kimi K2.6 UD Q8-K-XL (kinda 680Gb Ram 131k context) | Success✅️ | very good |
| release date | GLM 5.0 Q6-K-XL (650Gb Ram 131k context) | Success✅️ | --- |
| release date | Minimax M2.0 F16 GGUF (465Gb Ram 8k context) | Success✅️ | my quantization from original |
| release date | GLM 4.5 BF16 GGUF (712Gb Ram 8k context) | Success✅️ | same model in quality HQ4_K for ik_llama (366Gb Ram) and Q8 (392Gb Ram) also resolved test |
| release date | Deepseek V3.1 Q8 (720Gb Ram) | Success✅️ | resolved test but melody very primitive few sounds |
| release date | Deepseek V2.5-1210 Q8 (277Gb Ram) | Success✅️ | --- |
*Mozart test means - any Mozart melody in "ChucK" audio programming language, ChucK is very strict syntax language unforgivable to any mistakes. In 2026 from models like Kimi K2.6 and Deepseek V4 the test will be made harder by time, models of level and size of Kimi K2,6 and similar are capable to write full songs in ChucK in harmonic combination.
Incredible machine which provide so many for such little money, my best investment of just $1000 total(in 2024-25 prices, all except Gpu). During purchase i've planned upgrade with 128Gb DDR4 Ram modules, which i've heard produced, but on used market these are such rare that i've decided to finish machine by 64Gb modules, which was available quite in plenty places until November 2025. Its filled with Samsung and few Hynix each by 64Gb DDR4 ECC Ram. Ram is the hottest thing here, not GPU (4070 Ti Su), if no additional cooling made - RAM getting to 80C degrees immediately, my contraption cooling was able to cool Ram for usable 60C degrees (i've checked thoroughly with thermal camera). Some fans are hanging on cords to eliminate noise from vibration, thats the only way. This boards for newer CPUs they still sell on gigabyte.com/enterprise with support up to 24 Ram modules by user install.
This machine is final, can't see any replace for near future by golden Ram, only-GPU setups or Apple Macs i see as dead-end with no financial sense(no Ai model will return such expenses back on GPUs or Apple like ever, hobby should be the cheapest way possible).