AI Potluck
Model components / Fine-tuned / chat models

Qwen3-Coder

Alibaba Cloud

Qwen3-Coder-480B-A35B-Instruct (MoE, 480B total / 35B active, 160 experts, 256K native context to 1M w/ Yarn), non-thinking instruct/agentic-coding SKU; arXiv 2505.09388. HF repo created 2025-07-22.

Consolidated on 2026-07-29 from qwen2-5-coder-instruct; openness follows qwen3-coder, the release that currently governs. See docs/reference/identity.md. The earlier record dated the release 2025-05-14, which the card does not show; the repo's own creation date is 2025-07-22. Verified 2026-08-13 via the HF model card.

Openness

3 high confidence
3.0
weights
open(Apache-2.0, on HF)
data
closed(no training corpus released)
code
partial(inference/usage + chat template
license
Apache-2.0(OSI permissive, no use restrictions)

Apache-2.0 weights with a closed corpus and inference code only, and that holds across both generations - Qwen2.5-Coder and Qwen3-Coder are both plain Apache with no use restrictions. Qwen3-Coder is the current release, while the download volume still comes largely from Qwen2.5-Coder, which has had longer to accumulate it. Kept apart from the base Qwen tier because Alibaba ships and markets the Coder line separately.

  • https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct recorded 2026-08-13

    `license:apache-2.0` in the repo tags and `"gated":false` in the embedded repo state, so the BF16 weights download without a barrier; the card documents quickstart, tool-calling and agentic-coding usage only, with no post-training pipeline and no SFT/DPO/RL mixture named or released

Adoption

4 high confidence
4.0

The 32B-Instruct SKU alone reads 1,242,001 Hugging Face downloads in the trailing 30 days, which bands at level 4 by itself. Summed across the four declared SKUs the family reads 4,715,952 (32B-Instruct 1,242,001; 7B-Instruct 1,926,647; Qwen3-Coder-30B-A3B-Instruct 1,198,501; Qwen3-Coder-480B-A35B-Instruct-FP8 348,803), also level 4 - squarely inside the 1M-10M band on measured monthly download volume.

Capability

4 medium confidence
4.0

Strong open-weight coding model (Claude Sonnet-class agentic coding per vendor card), a notch below 2026 frontier coders (DeepSeek-V4-Pro/Kimi K2.6/GLM-5.1). HF card is vendor-reported; no independent SWE-bench number captured this run, so capped at 4 with medium confidence.

Verified 2026-08-13