Arcee Trinity Large Thinking 400B on NVIDIA Grace Blackwell Ultra GB300
Yes — Grace Blackwell Ultra GB300 runs Arcee Trinity Large Thinking 400B excellently at Q4_K_M — 41 tok/s. 288 GB VRAM with plenty of headroom.
Model Size
399B
Device VRAM
288 GB
Bandwidth
8000 GB/s
Quantization
Q4_K_M
Performance by Quantization
OwnRig currently has one published compatibility entry for Arcee Trinity Large Thinking 400B on NVIDIA Grace Blackwell Ultra GB300 at Q4_K_M. This is the best supported pairing we can stand behind today.
| Quantization | Speed | TTFT | Fits in VRAM | Rating | Confidence |
|---|---|---|---|---|---|
| Q4_K_M | 41 tok/s | 80ms | ✓ Yes | Excellent | estimated |
Notes
Q4_K_M
398B MoE (13B active). Q4_K_M ~255GB on 288GB discrete gpu. Q5_K_M (~295GB) exceeds 288GB VRAM.
About Arcee Trinity Large Thinking 400B
Arcee Trinity Large Thinking 400B (399B) is a chat, coding, ai coding, reasoning, multi-purpose model. Arcee AI's largest Mixture-of-Experts reasoning model. 399B total parameters with only 13B active per token (256 experts, 4 selected; 1.56% routing, one of the sparsest MoE architectures in production). 512K native context. Trained on 17 trillion tokens across 2,048 B300 GPUs. Post-trained with extended chain-of-thought reasoning and agentic reinforcement learning. Ranks #2 on PinchBench behind Claude Opus-4.6. Requires a GB300 (288 GB) for full in-VRAM inference; even the M4 Ultra 192 GB cannot fit Q3_K_M. US-built, Apache 2.0 licensed.
View all Arcee Trinity Large Thinking 400B hardware options →About NVIDIA Grace Blackwell Ultra GB300
NVIDIA Grace Blackwell Ultra GB300 has 288 GB at 8000 GB/s. Street price: $30,000.
See all models NVIDIA Grace Blackwell Ultra GB300 can run →Estimate method: Estimated from MoE architecture (active params per token), quantization size, and device bandwidth. Reference hardware source: huggingface.co (2026-03-14)
Performance varies by driver version, inference engine, quantization method, context length, and system configuration. Figures shown are estimates based on community benchmarks and may not reflect your exact setup. Product names are trademarks of their respective owners. OwnRig is independent and not affiliated with any hardware or AI model provider.