Swift Flash Next
#12 v Reasoning modelyUkisai · flash · od 21. September 2026 · 4× · naposledy 26. 9. 2026
39
Momentum
Swift Flash Next (Swift 1.5 Qwen3.8-Flash-Next) is a reasoning-efficient derivative of Qwen3.8-Flash-Next (MoE, 6B active parameters) developed by UkisAI. Its post-training targets shorter thinking traces along with improved coding, agentic and long-horizon task performance, while keeping accuracy close to the base model. It is available via an OpenAI-compatible API as well as BF16 weights and several quantized variants (GGUF, AWQ, NVFP4, AutoRound).
Vývoj momenta
28.06.26.09.
Vlastnosti
| Key Benchmark (%) | GPQA-Diamond: 89.6% at xhigh reasoning effort (trails base model by only 0.20 percentage points, with 63.4% fewer thinking tokens) |
| Context Window (Tokens) | 262,144 tokens |
| License | Swift Open License v1.0 (UkisAI contribution); base model under Qwen Community License 1.0. Free commercial use up to US$1,000,000 annual revenue, above that a Swift Enterprise License is required |
| Multimodality | Image-Text-to-Text (image and text input to text output); separate BF16 vision projector (0.91 GB) available |
| Platform | OpenAI-compatible API (ukisai.com/api/flash-next/v1), plus BF16 weights and GGUF/AWQ/NVFP4/AutoRound quantizations via Hugging Face |
| Price per 1M Tokens | No commercial list price found; API currently free for research purposes, no API key required |
| Release Date | September 21, 2026 |