[ccerv1]

What

This applies R-2026-11-08-m (recorded in #981): Qwen3-Coder-Next is another release of qwen-coder, and the earlier checkpoints are kept. No published score changes.

• Identity: Qwen/Qwen3-Coder-Next and -Base join the declared checkpoints; the four existing ones stay. The FP8 and GGUF builds of Next are left out as quantizations (R-2026-10-17-b). The description now covers the line.
• Openness: governing_release moves to qwen3-coder-next, the current release. It stays at 3, open weights: Apache-2.0, not gated, no training data or pipeline. check_rubric reproduces it. The axis is re-read and dated 2026-11-08 against Next's card, its Hub API record and the Qwen3-Coder README, which lists all three generations in one table. Artificial Analysis marks the 480B model deprecated in favor of Next.
• Adoption: the level stays 4, with no counts typed in. The old date was measured on the four earlier checkpoints only. So the axis is held with settled_by: scheduled_reconciliation, as fix(adoption): re-band colpali, openclip, siglip, modernvbert and tensorlake-sandbox on declared families and channels #904 did for colpali, until the next scheduled reconciliation measures the full declaration.
• Capability: the band is unchanged and stays held. The hold's reason now says the superseded-flagship question no longer applies under R-2026-11-08-m. What's still missing is an independent reading: Artificial Analysis publishes no Intelligence Index score for either release.

Follow-ups, not done here

• The 480B model is declared only as its FP8 build (Qwen3-Coder-480B-A35B-Instruct-FP8). That's a quantization; the BF16 checkpoint is undeclared. It was kept because the ruling says to retain the existing checkpoints.
• Several Qwen2.5-Coder checkpoints are undeclared: 0.5B, 1.5B, 3B and 14B Instruct, plus the base models. 14B-Instruct alone draws about 2M downloads a month. Declaring them would change the adoption sum, so it belongs in its own PR.

Checklist

• uv run python -m build.validate prints 0 error(s)
• New/changed scores cite sources (url, shows, accessed), each re-fetched 2026-11-08 with build.fetch_source
• I did not edit build/notebook_data.json or notebooks/ai-stack-map.py
• Product appears in exactly one category roster and one org roster: unchanged
• Label: none (hand work)

check_rubric, check_verification, check_capability and check_artifacts pass, as does preflight --skip-tests (all 18 locally runnable steps). Also run: the score-note and product-prose tests, and the identity, review-queue, adoption-freshness, verification, validate and rulings tests.

---

Generated by Claude Code