"Full" NVFP4 Qwen3.8 27B: Did you try this? I can't trust the eval. LiveCodeBench results for the BF16 baseline are 10~15 points below what they should be. Same for AIME. Evaluated with a low context length (32K), so in a condition where seeing quantization degradation is unlikely Just wondering whether I should spend one or two days on an RTX Pro 6000 to confirm, but at native context length.

Qwen3.8-27B NVFP4 variants are very close in accuracy. The real differences are memory and speed. If you're VRAM-limited, minima-ai/mnma_qwen3.8_27b_nvfp4 is a good pick, but it doesn't include MTP for faster inference. NVIDIA's version has MTP, but in my long-context coding tests MTP-4 is only ~2× faster than no MTP, and still ~2.5–3× slower than Unsloth (RTX Pro 6000). A likely reason: NVIDIA quantizes lm_head to NVFP4, while Unsloth keeps it FP8. Since MTP shares the target model's lm_head, this can hurt prediction quality and acceptance rate. So my pick is Unsloth NVFP4. Now testing accuracy on long-horizon agentic coding.

Sparkbench local battle: Qwen3.8-Flash-Next (NVFP4 TP2, spec off) scored 87.0 GLM-5.3-Flash (NVFP4 TP2, DFlash2) at 85.7 This is a deployment comparison, not a universal model ranking, but Qwen was faster, more reliable, and cleaner here. The most interesting result wasn’t the score: GLM hit four ~261K-token repetition loops, while Qwen’s NEXTN path had its own collapse and had to be disabled Local inference moves fast, but serving-stack correctness matters as much as the model. Full deployment and results:

Nvidia dropped an official DeepSeek-V4.1-Flash NVFP4 build. And this thing is BIG. DeepSeek V4.1 Flash stats 🧠 552B backbone 📚 +196B Engram conditional memory ⚡ only 8B active during prefill 🚀 16B active during decode 👁️ native vision 📖 1 MILLION token context 🧩 384 routed experts across 40 layers 📜 MIT license Nvidia has converted its routed MoE experts to NVFP4 W4A4 specifically for Blackwell GPUs. This is not a compressed giant model down to 4-bit. DeepSeek's experts were already stored in MXFP4, Nvidia instead converts them to its Blackwell-friendly NVFP4 format. And because NVFP4 uses finer scaling, the checkpoint actually gets slightly larger. 💾 Source: ~476 GiB 💾 NVIDIA NVFP4: ~492 GiB 48 safetensor shards. 😳 So why bother? Because NVIDIA is optimizing how those 4-bit experts execute on Blackwell. And impressively, NVIDIA's evaluations show basically no obvious quality collapse from the conversion. For example: 🧠 GPQA Diamond 91.04 → 91.29 💻 SciCode 54.40 → 55.84 🛠️ Terminal-Bench 2.1 81.60 → 82.16 👁️ MMMU-Pro 74.05 → 73.70 Some slightly up. Some slightly down. Essentially benchmark parity. And it already has: ✅ vLLM support ✅ SGLang support ✅ reasoning parser ✅ tool calling ✅ image input ✅ 1M context ⚠️ Nvidia validated it on 4× GB300 GPUs so not a local model (yet). The checkpoint is still ~492 GiB. 🔗 HF: /nvidia/DeepSeek-V4.1-Flash-NVFP4

🔥 GLM-5.3-Flash 320B-A18B MoE · NVFP4 · 4x DGX Spark TP4 Re-benched warmed + streaming: ⚡️ 63.8 tok/s PEAK on structured output (MTP runs hot) 🔢 ~61 median counting · 53 code · 37 prose 🧠 1.26M-token fp8 KV · 262K ctx · 0.2s TTFT Day-0. 320B-class reasoning on desk hardware 🖥️

🔥 The wildly popular DGX-Spark-friendly Qwen3.8-27B-NVFP4 now comes naked. The OrcaRouter family has already pulled 450K+ Hugging Face downloads across its FP8, GGUF, MLX, NVFP4 and BF16 builds. And now there’s a 23.4GB NVFP4 version built for NVIDIA Blackwell. 👀 🧠 Qwen3.8-27B dense 🚫 Abliterated / dramatically fewer refusals 👁️ Vision preserved 🛠️ Tool calling preserved 🚀 MTP preserved 📚 262K context 📦 ~23.4GB 📜 Apache 2.0 And this isn't a dumb quant conversion. It mixes precision ... ⚡ Most FFN layers → NVFP4 🧮 Attention + DeltaNet → FP8 🧠 Final FFN layers → FP8 💾 KV cache → FP8 👁️ Vision + embeddings + MTP → BF16 Target hardware 🔥 RTX 5090 🔥 DGX Spark / GB10 🔥 RTX PRO Blackwell 🔥 Multi-GPU RTX 50-series 🔥 B200 / B300 For AMD, Intel or Mac, there are also GGUF/MLX versions available 🔗 HF /orcarouter/Qwen3.8-27B-Uncensored-NVFP4

Benchmarking @NVIDIAAI's Nemotron Puzzle 75B locally on the GX10. NVFP4 via vLLM's OpenAI API, MTP speculative decoding, forced 1,500-token generations. 🏃‍♀️22.75 tok/s in a single session into 88.85 cumulative at 7 sessions. 🧍Baseline without MTP: ~16.3. Scripts + setup: