# Don't sleep on Nemotron 3.5 Super VL — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/74759971b569d236

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-19 04:01 ingested 2026-10-20 02:40

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1x9mwmz/dont_sleep_on_nemotron_35_super_vl/)

## retained text



```
Hey everyone,

One of the benchmarks I run for myself is to check obscure knowledge, like game rules from Anno 1602 (a game from 1998) and it's (many confusing) editions. Websearch is not permitted.

So far, it has come the closest for me in niche information.

Ranking order (1 highest, 6 lowest):

1. Nemotron 3.5 Super VL 120B-A12B: Hallucionated salt for citizen to merchant, left out church for settlers, left out required landmarks for merchant and aristocrat
2. Gemma 4 31B IT QAT: Omitted too many parts of the upgrade requirements
3. DeepSeek v4 Flash Vision Exp: Hallucionated different upgrade requirements
4. Qwen3.5-122B-A12B: Hallucionated game editions and upgrade requirements
5. Qwen3.8-flash-next: Hallucionated game editions and upgrade requirements
6. GPT-OSS 120B: Hallucionated game editions and upgrade requirements

Compared to the other models, that's a huge improvement! Basically from totally inaccurate to only minor mistakes.

The model itself (unsloth's quants, NVFP4) runs at \~8-10 t/s generation and \~150-170 t/s processing on dual RTX 5060 Ti 16GB with dual 48GB DDR5-6000MHz, no MTP, layer split, vulkan.

Seems like it's a good model to keep.  
Also crazy how capable Germma 4 31B IT QAT is.

\[EDIT\] inference engine and models used:

* llama.cpp 0.7.0, used with models: [https://github.com/ggml-org/llama.cpp/releases?page=4#release-b11669](https://github.com/ggml-org/llama.cpp/releases?page=4#release-b11669)
* MXFP4 for GPT-OSS-120B: [https://huggingface.co/bartowski/openai\_gpt-oss-20b-GGUF](https://huggingface.co/bartowski/openai_gpt-oss-20b-GGUF)
* Q4\_K\_S for Qwen3.5-122B-A10B: [https://huggingface.co/unsloth/Qwen3.5-122B-A10B-MTP-GGUF](https://huggingface.co/unsloth/Qwen3.5-122B-A10B-MTP-GGUF)
* Q4\_K\_XL for Qwen3.8-flash-next: [https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF](https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF)
* Q2\_K\_XL for DeepSeek v4 Flash Vision Exp: [https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF](https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF)
* QAT + MTP (Q4\_0) for Gemma 4 31B IT QAT: [https://huggingface.co/unsloth/gemma-4-31B-it-qat-GGUF](https://huggingface.co/unsloth/gemma-4-31B-it-qat-GGUF)
* NVFP4 for Nemotron 3.5 Super VL 120B-A12B: [https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF](https://huggingface.co/unsloth/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF)
```



document fetched 2026-10-20 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1x9mwmz/dont_sleep_on_nemotron_35_super_vl
