Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

Qwen · qwen3_next

Qwen3-Coder-Next-FP8

Qwen3-Coder-Next-FP8 by Qwen: a 79.7B-parameter mixture-of-experts open-weight model under the apache-2.0 license with tool calling.

apache-2.0Mixture of experts262K contextTool callingReasoningAuto-imported · not yet reviewed

1,423,926 downloads · 184 likes · Model card · synced 2026-09-19

Specifications

Parameters
79.7B (1.6B active)
Architecture
moe · qwen3_next
Layers
48
Hidden size
2,048
KV heads · head dim
2 · 256
Vocabulary
151,936
Native dtype
bfloat16
Context window
262,144 tokens
Released
2026-02-01
Last modified on Hub
2026-02-03

Features & licensing

License
apache-2.0
Commercial use
Yes
Modalities
text
Tool / function calling
Yes
Reasoning mode
Yes
Languages
EN
Quantised variants on Hub
FP8
Pipeline
text-generation
transformerssafetensorsqwen3_nexttext-generationconversationalendpoints_compatiblefp8

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run Qwen3-Coder-Next-FP8

Weights need about 175 GB at FP16, 92 GB at INT8 and 53 GB at INT4. The KV cache adds roughly 98 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 71.6 GB, which fits on 1 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
4 × NVIDIA L4 (24 GB)96 GB75 %~0.5 s$2,336
2 × NVIDIA L40S (48 GB)96 GB75 %~0.2 s$2,774
1 × NVIDIA A100 (80 GB)recommended80 GB90 %~0.1 s$2,336
1 × NVIDIA H100 (80 GB)80 GB90 %~0.0 s$3,285

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.