Mixture of Experts: upcoming dates

Recent

• AI10/12/2026@TechDesk

  Mistral Large 4 Released

  15 Rue des Halles, 75001 Paris, FranceMistral AI

  

  Starts 10/06/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins84% CONFIDENCEOverall confidence: 84%How well the pin's source and references back up its dates.Weighted average of how firmly 5 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better12 days ago

  Mistral's^[1][5] announcement 'Introducing Mistral Large 4' is dated October 6, 2026 and describes a public preview; its docs model card lists the release date as October 6, 2026 (v26.10, Public Preview) and TechCrunch's^[2] report is dated the same day.

  Mistral AI releases Mistral Large 4 in public preview, a natively multimodal mixture-of-experts model of about 1 trillion parameters trained on its own cluster, with open weights promised by the end of October 2026.

  80
• AI10/12/2026@TechDesk

  Inkling-Small Released

  2300 Harrison Street, San Francisco, California 94110, United StatesThinking Machines Lab

  

  Starts 07/30/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins88% CONFIDENCEOverall confidence: 88%How well the pin's source and references back up its dates.Weighted average of how firmly 4 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better80 days ago

  Thinking Machines^[1][4]' post "Introducing Inkling-Small" is dated Jul 30, 2026 and opens "Today, we are releasing Inkling-Small"; the original Inkling post's update note says it launched "two weeks after this post was published" and OpenRouter^[3] lists it as released Jul 30, 2026.

  Thinking Machines Lab releases Inkling-Small, an open-weights 276B-parameter model with 12B active parameters that matches or beats the 975B Inkling on reasoning and agentic coding benchmarks, including 80.2% on SWE-bench Verified.

  80
• AI10/12/2026@TechDesk

  Inkling Released

  2300 Harrison Street, San Francisco, California 94110, United StatesThinking Machines Lab

  

  Starts 07/15/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins87% CONFIDENCEOverall confidence: 87%How well the pin's source and references back up its dates.Weighted average of how firmly 5 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better95 days ago

  Thinking Machines^[1][3]' post "Inkling: Our Open-Weights Model" is dated Jul 15, 2026 and says "Today we are advancing our mission by releasing a model we trained from scratch"; The Next Web^[5] ran its report on July 15 and OpenRouter^[4] lists it from Jul 17.

  Thinking Machines Lab releases Inkling, its first model and an open-weights 975B-parameter mixture-of-experts system (41B active) with a 1M-token context that reasons over text, images and audio, built as a base for fine-tuning.

  100
• AI10/12/2026@TechDesk

  Trinity-Large-Thinking Released

  San Francisco, California, United StatesArcee AI

  

  Starts 04/01/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins84% CONFIDENCEOverall confidence: 84%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better200 days ago

  Arcee's^[1][2] post "Trinity-Large-Thinking: Scaling an Open Source Frontier Agent" is dated Apr 1, 2026 and says "Today we are releasing Trinity-Large-Thinking on our API and the weights on Hugging Face^[4]"; OpenRouter^[5] lists it as released Apr 1, 2026.

  Arcee AI releases Trinity-Large-Thinking, a 398B-parameter open reasoning model (about 13B active) built for long-horizon agents, which it says ranks second on PinchBench behind Claude Opus 4.6 at $0.90 per 1 million output tokens.

  140
• AI10/12/2026@TechDesk

  Qwen3.5-122B-A10B, Qwen3.5-35B-A3B and Qwen3.5-27B Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 02/24/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins85% CONFIDENCEOverall confidence: 85%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better236 days ago

  The Hugging Face^[1][2] repositories for Qwen3.5-122B-A10B, -35B-A3B and -27B were completed on 24 February 2026 (UTC; commits 13:43-13:58 for 122B-A10B and 27B, 11:20 for 35B-A3B) and Novita AI dates the release 24 February; VentureBeat^[5] (25 Feb, 7:39 pm PT), Gigazine^[4], EqualOcean and OpenRouter^[6] give 25 February, the same launch seen from China and US time.

  Alibaba's Qwen team releases three open-weight Qwen3.5 mid-size models under Apache 2.0, a 122B mixture-of-experts with 10B active, a 35B with 3B active and a 27B dense model, which it says beat OpenAI's GPT-5 mini on most benchmarks.

  60
• AI10/12/2026@TechDesk

  Qwen3.5-397B-A17B and Qwen3.5-Plus Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 02/16/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins87% CONFIDENCEOverall confidence: 87%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better244 days ago

  CNBC^[5], published Monday 16 February 2026, reports that Alibaba's open-weight and hosted Qwen3.5 models "were made available on Monday, the eve of the Chinese New Year"; OpenRouter^[6] lists Qwen3.5 397B A17B as released Feb 16, 2026. Qwen's^[1] blog page is stamped 2026/02/15 and the Alibaba Cloud^[4] Community repost 17 February.

  Alibaba releases Qwen3.5-397B-A17B, an open-weight native multimodal model that activates 17B of 397B parameters, plus the hosted Qwen3.5-Plus with a 1M-token context, on the eve of Lunar New Year.

  70
• AI10/12/2026@TechDesk

  Trinity Large Preview, Base and TrueBase Released

  San Francisco, California, United StatesArcee AI

  

  Starts 01/27/2026CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins88% CONFIDENCEOverall confidence: 88%How well the pin's source and references back up its dates.Weighted average of how firmly 4 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better264 days ago

  Arcee's^[1][2] post "Trinity Large: An Open 400B Sparse MoE Model" is dated Jan 27, 2026 and says "We're putting out three variants"; the Trinity-Large-TrueBase, Base and Preview repositories on Hugging Face^[4] were all created that day.

  Arcee AI releases Trinity Large, a 400B-parameter sparse mixture-of-experts model with 13B active, as a chat-ready Preview plus Base and TrueBase checkpoints from a 17T-token run it says cost $20 million.

  80
• AI10/12/2026@TechDesk

  MiniMax-M2.1 Released

  Xinyan Mansion, Guiqing Road, Xuhui District, Shanghai, ChinaMiniMax

  

  Starts 12/23/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins84% CONFIDENCEOverall confidence: 84%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better299 days ago

  MiniMax's^[1] news post is dated 2025.12.23 and says the M2.1 API is now live on its Open Platform and the weights are open-source; OpenRouter^[5] lists minimax/minimax-m2.1 on 23 December 2025 and Artificial Analysis^[4] gives 23 December 2025, while the Hugging Face^[6] repository was created on 20 December and its weights uploaded on 21 December.

  MiniMax releases M2.1, an open-weight 230B-parameter mixture-of-experts coding and agent model that activates 10B parameters and scores 74.0 on SWE-bench Verified and 88.6 on its new VIBE full-stack benchmark.

  80
• AI10/12/2026@TechDesk

  MiniMax-M2 Released

  Xinyan Mansion, Guiqing Road, Xuhui District, Shanghai, ChinaMiniMax

  

  Starts 10/27/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins86% CONFIDENCEOverall confidence: 86%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better356 days ago

  MiniMax's^[1] news post is dated 2025.10.27 and announces M2 as open-sourced with its API and MiniMax Agent free for a limited time; VentureBeat^[3] covered it at 12:01 pm PT on 27 October, while the Hugging Face^[5] weights were uploaded on 25 October UTC and Artificial Analysis^[4] gives 26 October.

  MiniMax open-sources M2, a 230B-parameter mixture-of-experts model with 10B active parameters built for coding and agentic work, priced at $0.30 per million input tokens and ranked first among open-weight models on Artificial Analysis's Intelligence Index at launch.

  120
• AI10/12/2026@TechDesk

  Qwen3-VL-235B-A22B-Instruct and Qwen3-VL-235B-A22B-Thinking Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 09/23/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins86% CONFIDENCEOverall confidence: 86%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.1 years ago

  Simon Willison's^[5] post of 23 September 2025 quotes Alibaba's "Firstly, we are open-sourcing the flagship model of this series"; the Hugging Face^[1] weights for the Instruct (05:25Z) and Thinking (04:16Z) repositories finished uploading on 23 September and OpenRouter^[6] lists the Instruct model at 23 September, with The Decoder^[4] reporting it on 25 September.

  Alibaba open-sources the flagship Qwen3-VL-235B-A22B in Instruct and Thinking versions, a 235B vision-language mixture-of-experts with a 256K-token native context that Alibaba says matches or exceeds Gemini 2.5 Pro on major visual-perception benchmarks.

  90
• AI10/12/2026@TechDesk

  Qwen3-Next-80B-A3B-Instruct and Qwen3-Next-80B-A3B-Thinking Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 09/11/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins89% CONFIDENCEOverall confidence: 89%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.1 years ago

  AINews's issue for 10-11 September 2025 calls it "Today's Qwen3-Next release"; OpenRouter^[6] lists both models as "Released Sep 11, 2025" and Qwen's^[1] launch post on X went out at 17:50 UTC on 11 September (Qwen's blog page is stamped 10 September, and Simon Willison^[4] and Asia press covered it on 12 September).

  Alibaba releases Qwen3-Next-80B-A3B-Instruct and Thinking, ultra-sparse 80B models that activate only 3B parameters per token, run a hybrid Gated DeltaNet attention design and carry an Apache 2.0 licence.

  50
• AI10/12/2026@TechDesk

  Qwen3-30B-A3B-Instruct-2507 Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 07/29/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins85% CONFIDENCEOverall confidence: 85%How well the pin's source and references back up its dates.Weighted average of how firmly 5 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.2 years ago

  Simon Willison's^[4] post dated 29 July 2025 quotes Qwen's tweet announcing Qwen3-30B-A3B-Instruct-2507 and runs it that day, and OpenRouter^[5] lists it as "Released Jul 29, 2025"; Qwen's own repository log dates it 2025.07.30 (the Beijing date), so sources differ by one day.

  Alibaba's Qwen team releases Qwen3-30B-A3B-Instruct-2507, a non-thinking 30.5B mixture-of-experts model with only 3.3B active parameters and a 262K context that Qwen says approaches GPT-4o.

  110
• AI10/12/2026@TechDesk

  Qwen3-235B-A22B-Thinking-2507 Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 07/25/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins85% CONFIDENCEOverall confidence: 85%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.2 years ago

  VentureBeat's^[4] 25 July 2025 article reports the release that day; the Qwen3 repository's news log reads "2025.07.25: We released the updated version of Qwen3-235B-A22B thinking mode, named Qwen3-235B-A22B-Thinking-2507", and OpenRouter^[6] lists it as released Jul 25, 2025.

  Alibaba's Qwen team releases Qwen3-235B-A22B-Thinking-2507, a thinking-only 235B mixture-of-experts model with a 262K context under Apache 2.0, scoring 92.3 on AIME25 and 74.1 on LiveCodeBench v6.

  70
• AI10/12/2026@TechDesk

  Qwen3-235B-A22B-Instruct-2507 Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 07/21/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins85% CONFIDENCEOverall confidence: 85%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.2 years ago

  Qwen's Qwen3 repository news log reads "2025.07.21: We released the updated version of Qwen3-235B-A22B non-thinking mode, named Qwen3-235B-A22B-Instruct-2507"; Simon Willison's^[5] 22 July post calls it "published yesterday without much fanfare" and OpenRouter^[6] lists it as released Jul 21, 2025.

  Alibaba's Qwen team quietly releases Qwen3-235B-A22B-Instruct-2507, a non-thinking 235B mixture-of-experts model with a 256K context under Apache 2.0 that Qwen's own tables put ahead of Claude Opus 4 non-thinking and Kimi K2 on several tests.

  70
• AI10/12/2026@TechDesk

  MiniMax-M1 Released

  Xinyan Mansion, Guiqing Road, Xuhui District, Shanghai, ChinaMiniMax

  

  Starts 06/16/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins86% CONFIDENCEOverall confidence: 86%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.3 years ago

  MiniMax's^[1] news post is dated 06.16.2025 and its technical report was submitted to arXiv^[6] on June 16, 2025; Analytics Vidhya^[3] (18 June) says M1 was released on day 1 of MiniMaxWeek and OpenRouter^[4] lists minimax/minimax-m1 on 17 June (UTC).

  MiniMax releases M1, an open-weight 456B hybrid mixture-of-experts reasoning model with lightning attention, a 1M-token context and 80K-token reasoning output, trained with a new reinforcement-learning algorithm for a reported $534,700.

  10
• AI10/12/2026@TechDesk

  Qwen3-32B and the Qwen3 Family Released

  Wuchang, Yuhang District, Hangzhou City, Zhejiang, ChinaAlibaba Cloud

  

  Starts 04/29/2025CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins89% CONFIDENCEOverall confidence: 89%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better1.5 years ago

  Qwen's blog post is dated April 29, 2025 and opens "Today, we are excited to announce the release of Qwen3"; Alibaba Group's^[3] English release is dated April 29, 2025, TechCrunch^[4] published at 2:37 PM PDT on April 28 and OpenRouter^[5] lists Qwen3 32B at 28 April (UTC), the same launch seen from US time.

  Alibaba's Qwen team open-sources the Qwen3 family under Apache 2.0, six dense models from 0.6B to 32B and two mixture-of-experts models (30B-A3B and 235B-A22B), with switchable thinking and non-thinking modes and 119 languages; the dense flagship Qwen3-32B scores 93.8 on Arena-Hard in thinking mode.

  130
• AI09/30/2026@TechDesk

  Gemini 1.5 Pro Released

  Googleplex, 1600 Amphitheatre Parkway, Mountain View, California 94043, United StatesGoogle

  

  Starts 02/15/2024CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins82% CONFIDENCEOverall confidence: 82%How well the pin's source and references back up its dates.Weighted average of how firmly 2 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better2.7 years ago

  Google's post "Our next-generation model: Gemini 1.5" is dated 15 February 2024 and opens a limited preview of 1.5 Pro for developers and enterprise customers that day.^[1]

  Google unveils Gemini 1.5 Pro, a mixture-of-experts model that matches Gemini 1.0 Ultra and introduces a context window of up to 1 million tokens, tested to 10 million.

  90