Qwen3.8-Flash-Next @ GB x 2 running ~74 tok/s for code 🤯
While for thinking and code it gets ~59 tok/s 🚀
Best V2 recipe just got improved and got 2.85M kv FP8 🥳
⚡️https://t.co/MyV5S3BcTE
This is for you single DGX Spark owners 💫
You can now run Qwen3.8 Flash NVFP4 on one DGX Spark with great performance!
- Up to 1M context, 1,431,164 KV cache
- Full image & video support
- 37 decode tok/s on prose single stream
- Up to 86 tok/s on prose 4 concurrent streams
- 1500-2000 tok/s prefill on any size.
- Stress tested with 400k prefill - passed.
IMO this is the BEST model to run now on a single dgx spark. It's better than Qwen3.8-27B, and also better than the 1x dgx spark version of DeepSeek v4 Flash.
Get it here:
https://t.co/LG0I9PJKxI
🚨🔴 GABRIEL MARTINELLI LEAVES ARSENAL FOR AL HILAL. 🇸🇦
A chapter closes.
From that fearless kid on the left wing to the player who gave us countless moments of joy and a Premier League title, Martinelli leaves with our gratitude.
Thank you for everything, Gabi. ❤️
Go well, Starboy. 🌟🔴⚪️
After a lot of testing, crashes, tuning, and DFlash2 work, I’ve got Qwen 3.8 27B NVFP4 running at 50+ tokens/sec on a single DGX Spark.
For me, this is now the best all-round model to run on one Spark: fast, capable, long-context ready, and genuinely usable for agent and coding workloads.
Full reproducible recipe:
https://t.co/O587Uxynk5
NVIDIA just nuke 30 years of animation technology.
They open-sourced MotionBricks.. one neural model that generates 350,000+ character motions in real time.
→ 5,000 FPS
→ 2ms latency
→ Zero-shot to new tasks
→ Drops straight into Unreal Engine 5
No foot-locking. No blending. No collision detection. No hand-authored transitions.
100% Open Source.
Great way to see how each model runs. Keep in mind GLM-5.3 Flash will only improve from here.
Speed:
DS4F > Qwen3.8 Flash > GLM-5.3 Flash
Intelligence:
GLM-5.3 Flash > Qwen3.8 Flash > DS4F
Qwen3.8 Flash might be a good balanced choice if you want a combination of both speed and intelligence. It also has full vision, which DS4F lacks.
Link to all recipes:
Qwen3.8-Flash
https://t.co/e4WwEShW25
GLM-5.3 Flash
https://t.co/6o0dNhWANq
DeepSeek v4 Flash
https://t.co/6HXa9pxY6R
sparkDash:
https://t.co/8H3h5wknDP