Open-weight model release tracker: upcoming and recent drops

Checked: October 25, 2026. This page answers one question: when do the weights actually drop? Every date and download link comes from an official lab post or Hugging Face model card. Licenses are marked Unknown until we open the official license text (or the HF card states it clearly). We do not host weights. For day-to-day lab news, see the AI Lab Tracker. This page has no affiliate links.

TL;DR

| Status | Model | Weights date (official) | License (as stated) | |---|---|---|---| | Upcoming | Kimi K3.1 Pro | By Nov 4, 2026 | Unknown (not published yet) | | Upcoming | Mistral Large 4 | End of October 2026 (promised) | Unknown until release | | Upcoming | MiniMax M3.1 Flash | "Later" (no hard date on official post) | Unknown | | Out | DeepSeek-V4.1-Flash-1017 | Oct 17, 2026 | MIT (HF card) | | Out | Muse Glimmer 30B | Oct 19, 2026 (Meta research post) / HF card | Apache 2.0 (HF card) | | Out | Muse Spark Open | Oct 15, 2026 | Muse Model License (limited commercial terms; not OSI) |

---

Upcoming (watch list)

Kimi K3.1 Pro — full weights by Nov 4, 2026

Moonshot's K3.1 Pro is live on Kimi.ai, Kimi Work, Kimi Code and the Kimi API today. The official tech blog states: "Full model weights will be released by Nov 4, 2026." The same wording appears on the Kimi API quickstart. Weights are not on Hugging Face yet as of this check.

Size (vendor): 2.8T total / 104B active, 1M context, text + image input.
API list price (vendor): $2 input / $10 output / $0.20 cache hit per M tokens.
Terminal-Bench 4.0: Moonshot reports 65.7%; Artificial Analysis (third-party) reported 67%. Do not mix the two.
License: Unknown until the weight release ships a LICENSE file.
Official: Kimi K3.1 Pro tech blog · API quickstart

Mistral Large 4 — open weights promised by end of October

Mistral's Large 4 announcement (Oct 6, 2026) describes a 1T-parameter (52B active) multimodal MoE in public preview and says open weights are coming by the end of October. As of Oct 25 we have not confirmed a Hugging Face upload for Large 4 itself. Treat the date as a promise, not a completed drop.

License: Unknown until release.
Official: mistral.ai/news/mistral-large-4

MiniMax M3.1 Flash — weights "later"

MiniMax's Oct 19 post prices the API at $0.15 / $0.60 per M tokens ($0.015 cached) and says weights will be released later. A third-party summary claimed Oct 23; that date is Unverified against the official post, so we do not list it as a deadline.

License: Unknown.
Official: minimax.io/blog/minimax-m31f

Also watching (no public weights yet)

Block Dessa — full-duplex speech-to-speech research preview. Block says it can run locally on a MacBook Pro M4 Max or better, but has announced no public API, weights or pricing. block.xyz/dessa

---

Recently released (weights available)

DeepSeek-V4.1-Flash-1017 — MIT, on Hugging Face

Minor update to V4.1-Flash (Oct 17, 2026). The Hugging Face model card states: "This repository and the model weights are licensed under the MIT License."

Download: huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
Vendor Terminal-Bench 4.0: 42.3 (prior Flash 31.2). Artificial Analysis (third-party) lists Intelligence Index 45 and calls it a top open-weights model — that ranking is third-party, not DeepSeek's claim.
Docs/changelog: api-docs.deepseek.com/updates

Muse Glimmer 30B — Apache 2.0

Meta's ~30B on-device agentic multimodal model. The Hugging Face model card lists License: Apache 2.0 and ships BF16 safetensors plus quantized builds.

Download: huggingface.co/meta-models/Muse-Glimmer-30B
Meta docs: dev.meta.ai/docs/muse-glimmer
Tracker research URL used on our Lab Tracker: ai.meta.com research note

Muse Spark Open — Muse Model License (not OSI)

Open weights released Oct 15, 2026: 816B total / 42B active, 1M context, multimodal. The HF card and Meta materials put it under the Muse Model License, which has limited commercial terms. Do not treat this as Apache/MIT-style open source.

Download: huggingface.co/meta-models/Muse-Spark-Open-816B-A42B
Official blog (Lab Tracker URL): research.meta.ai/blog/muse-spark-open

---

How we keep this list honest

1. Official date or it doesn't go in the "Upcoming" deadline column. Blog wording like "later" stays as "later". 2. License = Unknown until the weight repo or an official license page says otherwise. "Open weights" ≠ OSI open source (see Muse Spark Open). 3. Benchmarks are labelled vendor-reported or third-party. We never invent scores or claim we ran the evals. 4. Cross-link the AI Lab Tracker for the broader day-to-day feed; this page is only the weight-drop landing page.

If a lab posts a new weight drop and we miss it, the next daily update pass should catch it from /workspace/news/ and the Lab Tracker.

---

Sources (checked 2026-10-25)

1. Moonshot — Kimi K3.1 Pro tech blog (weights by Nov 4, 2026): https://www.kimi.ai/blog/kimi-k3-1-pro 2. Moonshot — Kimi K3.1 Pro API quickstart (same weights wording): https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart 3. DeepSeek — V4.1-Flash-1017 on Hugging Face (MIT): https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 4. Meta — Muse Glimmer 30B on Hugging Face (Apache 2.0): https://huggingface.co/meta-models/Muse-Glimmer-30B 5. Meta — Muse Spark Open on Hugging Face (Muse Model License): https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B 6. Mistral — Large 4 announcement (open weights by end of October): https://mistral.ai/news/mistral-large-4/ 7. MiniMax — M3.1 Flash blog (weights later): https://www.minimax.io/blog/minimax-m31f 8. Block — Dessa product page: https://block.xyz/dessa 9. StackLoom — AI Lab Tracker: https://stackloom-ai.pages.dev/tracker/