Qwen3.8-27B
Alibaba · Sorti août 2026
Qwen3.8-27B is Alibaba's 27-billion-parameter open-weight model announced August 3, 2026 as the single-GPU companion to the 2.4T Qwen3.8-Max flagship. Open weights were promised for the week of August 10, 2026 on Hugging Face and ModelScope but have not yet shipped. No benchmarks, no license, no architecture details, and no context window have been published — the model is a commitment with a parameter count. If it ships as promised under a permissive license, it could become the most important local AI release of the year; until then, it's vaporware backed by Alibaba's word.
Est-ce fait pour vous ?
Recommandé pour
- Single-GPU self-hosting — 27B parameters fits a single consumer or datacenter GPU when quantized (4-bit: ~15GB VRAM, FP8: ~27GB, BF16: ~54GB)
- Open-weight ownership — full control over fine-tuning, data privacy, and deployment without API vendor lock-in
- Qwen 3.6 lineage — predecessor Qwen3.6-27B delivered strong open-weight coding and agent performance under Apache 2.0; 3.8 generation inherits training improvements
- Cost-sensitive volume deployments — self-hosting breaks even vs. API at several billion tokens/month
Déconseillé pour
- Immediate production use — no benchmarks published, no license disclosed, weights not yet on Hugging Face
- Regulated environments requiring auditable model cards, verified safety evaluations, and clear license terms
- Maximum capability workloads — the 27B model will not match the 2.4T Qwen3.8-Max flagship on complex reasoning or frontier benchmarks
- Teams who need predictable API pricing and SLAs — no API offering exists; this is a self-host-only model
Tarifs
Entrée
Free (open weights — self-hosted)
Sortie
Free (open weights — self-hosted)
Contexte
TBD — context window not published
Aucun changement de verdict pour l’instant
L’horloge tourne dès le premier jour : les changements apparaîtront ici à mesure que notre verdict évolue.
Sources
- Hugging Face — huginnfork/Qwen3.8-27B-FP8 placeholder (confirms no weights yet)août 2026
- Swfte AI — Qwen3.8-27B: The Version You Can Actually Run (Aug 4, 2026)août 2026
- AI FrontPage — Alibaba Launches Most Capable Qwen3.8-Max (Aug 3, 2026)août 2026
- DataNorth — Alibaba releases Qwen3.8-Max (Aug 4, 2026)août 2026
- MarkTechPost — Alibaba Qwen Releases Qwen3.8-Max (Aug 3, 2026)août 2026
- YottaLabs — Qwen 3.8 27B: Specs, Hardware Requirements, and How to Run It (Aug 4, 2026)août 2026
- Digital Applied — Qwen3.8 Open Weights: Check This Before Downloading (Aug 7)août 2026
Journal de vérification
Aucune vérification pour le moment
Nous n’avons pas encore enregistré de vérification pour cette entrée. Dès qu’une vérification s’exécute, son historique apparaît ici.