Channel
@sinceaihq
On this record: Topic · Growth · Engagement · Reactions · Posts · Cite this entry
237subscribers
+89 since we began measuring on 7 August 2026
Risers and fallers across the register · movement among entries of Under 1,000.
| Telegram ID | -1003422351784 |
|---|---|
| Type | Channel |
| Username | @sinceaihq |
| Created | Between 1 November 2025 and 31 March 2026 — estimated from Telegram’s id allocation, not measured. How this range is calculated. |
| First recorded | 8 August 2026 |
| Last confirmed live | 10 October 2026 |
| Measurements held | 10 |
| Confirmed unchanged | 1 time, most recently 10 October 2026 |
| On Telegram | t.me/sinceaihq |
Technology — a classification, not a measurement. An on-box language model (Qwen3.6-35B-A3B-FP8, prompt version 1) read this channel’s own recent posts on 24 September 2026 and assigned it the closest of 31 fixed categories, at 100% confidence. This is a model’s judgement about what the channel is likely to be about, not a fact this register measured the way a subscriber count or a view count is measured — it can be revised on a later pass, and it carries no weight anywhere else on this page. How this classification works, and why it has no browse page of its own yet.
| Measured (UTC) | Subscribers | Change |
|---|---|---|
| 10 Oct 2026, 15:56 | 237 | +18 |
| 30 Sept 2026, 03:00 | 219 | +27 |
| 16 Sept 2026, 06:16 | 192 | +10 |
| 8 Sept 2026, 13:35 | 182 | +10 |
| 30 Aug 2026, 20:24 | 172 | +6 |
| 23 Aug 2026, 07:15 | 166 | +11 |
| 15 Aug 2026, 15:23 | 155 | +5 |
| 8 Aug 2026, 20:15 | 150 | no change |
| 8 Aug 2026, 20:12 | 150 | +2 |
| 7 Aug 2026, 23:31 | 148 | first reading |
20 posts held, back to 1 July 2026 — the reader has not yet reached the start of this channel’s public history, so older posts may sit further back, unread. Read across 1 page of Telegram’s post history, 20 posts per page.
Nothing published in the last 30 days. ERR and ER are rolling 30-day measures, so there is nothing to compute — we hold 20 posts for this entry, the most recent from 8 August 2026. An engagement rate over an empty window would be a number about nothing.
94 reactions across 20 posts, in 3 distinct kinds. The most used accounts for 87.2% of them.
| Reaction | Count | Share | Share, drawn |
|---|---|---|---|
| 🔥 | 82 | 87.2% | |
| 👍 | 9 | 9.57% | |
| ❤ | 3 | 3.19% |
No sentiment is inferred, and none should be read in. This table is ordered by count and by nothing else. Emoji do not carry stable meaning across languages or communities — 🙏 is thanks in one channel and mourning in another — so we publish which ones were pressed and how often, and pass no judgement on what an audience meant by them.
Precision. Telegram publishes reaction counts per emoji and short-forms each one — 4.34K, 1.2M — so any single kind at or above 1,000 reaches us at three significant figures, and only counts below 1,000 are exact. The shares above are ratios of those figures and carry the same error. This is also why the total here can differ slightly from a reaction total printed elsewhere on the page: both are sums of the same rounded parts, taken over samples with different edges.
Coverage. Reactions were read on 20 of the 20 sampled posts in this sample. Summed by Telegram’s own count on each post — not by adding up the per-emoji breakdown above — those same posts carry 94 reactions in total: the kind of figure the paragraph above means by “a reaction total printed elsewhere on the page”.
Measured over the 20 most recent posts we hold, published 1 July 2026 to 8 August 2026, using the newest reading held for each. Telegram Stars are excluded: they are a payment, not a reaction, and they have their own section.
🧠 For AI researchers and agent developers, Microsoft Research just open-sourced Orchard for training and evaluating scalable agentic AI systems Orchard provides reusable Kubernetes-based environments for software engineering, web navigation, and personal-assistant tasks, with releases including Orchard-SWE, Orchard-GUI, and Orchard-Claw. The framework includes training recipes, datasets, and evaluation methods desig…
🔥3
🤖 For AI developers and researchers, Alibaba just launched Qwen3.8-Max, a 2.4T-parameter multimodal MoE model with a 1M-token context window Qwen3.8-Max uses a sparse Mixture-of-Experts architecture with 95 billion active parameters per inference step, making it Alibaba’s largest flagship model to date. The model is designed for long-horizon agentic workflows, autonomous coding, and multimodal reasoning workloads. A…
🔥3👍2
⚙️ For AI infrastructure teams training MoE models on NVIDIA NVL72s, Cursor just open-sourced a fused communication+compute megakernel Mixture-of-Kittens (MoK) is a production MoE training kernel built for NVIDIA GB300 NVL72 systems that combines MoE communication and computation into a single deterministic kernel. Cursor reports up to 2.37× higher MXFP8 forward throughput versus public baselines and a 1.41× end-to-…
🔥5
🔬 For LLM researchers and open-model builders, VIDRAFT just released a fully open 6.59B-parameter MoE model designed to study heterogeneous attention placement Aether-7B-5Attn ships with training code, checkpoints, logs, architecture details, and a reproducible data recipe under an Apache-2.0 license. The model distributes five different attention mechanisms across layers using a Latin-square design, creating a cont…
🔥6
🧠 For AI researchers building long-context models, Sparse Delta Memory scales RNN state sizes by orders of magnitude without increasing per-token compute Researchers from Meta FAIR and collaborators introduced Sparse Delta Memory (SDM), a sparse-memory extension of Gated DeltaNet designed to expand recurrent model memory efficiently. The paper reports stronger long-context retrieval, in-context learning, and scaling…
🔥4
🌕 For AI developers and researchers, Moonshot AI just released Kimi K3 — a 2.8T-parameter open-source model built to compete with top U.S. systems Kimi K3 ships with a 1 million-token context window, multimodal support, and OpenAI SDK compatibility, alongside benchmark results showing strong performance on coding, agentic workflows, and long-horizon reasoning tasks. The release also introduces new attention architec…
🔥5
🧠 For multimodal AI developers, Thinking Machines Lab just released a 975B-parameter open-weights MoE model with native text, image, and audio reasoning Inkling uses 41B active parameters and supports context windows up to 1M tokens. The Apache 2.0 release includes Hugging Face weights plus integrations with transformers, vLLM, SGLang, llama.cpp, and major inference providers. Thinking Machines Lab also published be…
🔥2👍1
🔬 For computational biology, chemistry, and materials science researchers, SciReasoner introduces a single foundation model for structure-aware reasoning across proteins, molecules, and crystals Researchers from Shanghai AI Laboratory and collaborators built a multimodal model that uses unified structural tokenization and autoregressive reasoning traces tied to explicit scientific evidence. SciReasoner achieved stat…
🔥7
🇩🇪 For AI teams building bilingual LLMs, Germany’s Soofi S 30B just raised the bar for fully open models The KI Bundesverband-led consortium released Soofi S 30B-A3B, an open-weight mixture-of-experts model trained entirely on Deutsche Telekom’s AI cloud infrastructure. It combines a hybrid Mamba-Transformer architecture with only 3.2B active parameters per token, enabling stable throughput at long context lengths. …
🔥6
🤖 For AI teams building agents, OpenAI just released the GPT-5.6 model family with native multi-agent orchestration and programmatic tool calling The launch includes the flagship Sol model alongside lower-cost Terra and Luna variants, all focused on coding agents, enterprise knowledge workflows, long-context reasoning, and advanced tool use. OpenAI also added Programmatic Tool Calling and multi-agent support to the …
🔥5
🤖 For AI engineers building agents and coding tools, Tencent just released a smaller-footprint open model aimed at competing with larger flagships Tencent’s new Hy3 model uses a 295B-parameter Mixture-of-Experts architecture with only 21B active parameters, alongside a 256K context window and Apache 2.0 licensing. The company is positioning Hy3 around efficiency and deployability rather than maximum scale, with benc…
🔥4
🎧 For multimodal AI engineers, NVIDIA just released a 30B audio-text LLM that keeps near-parity with its text backbone on reasoning benchmarks Audex (Nemotron-Labs-Audex-30B-A3B) uses a unified decoder architecture for both audio understanding and generation while remaining compatible with standard LLM tooling like Megatron-LM and vLLM. The mixture-of-experts model supports context windows up to 1M tokens and is one…
🔥5
Showing the 12 most recent of 20 posts we hold for @sinceaihq. View and reaction counts are the latest single reading for each post, not a live figure, and a recent post is still accumulating both. A view count marked ≈ was rounded by Telegram before we ever saw it — t.me prints views in full below 1,000 and to three significant figures above, so ≈1,200,000 means somewhere between 1,150,000 and 1,249,999. Unmarked counts are exact. Text is reproduced from the public post preview and truncated for length.
A live page changes as we take new readings, so a citation should name the measurement it is based on, not just the URL. The line below cites the subscriber count as measured 10 October 2026 — this entry's latest reading, not the date you are reading this.
“Since AI” (@sinceaihq), 237 subscribers as measured 10 October 2026. Telegram Register, tgregister.com/channel/sinceaihq.
Full measurement history, CC BY 4.0. Every reading this register holds for this entry, not just the latest one, as a dated, downloadable record: CSV · JSON. Free to use with attribution to tgregister.com. Each file carries its own generation timestamp, which is the figure to cite for exactly when the data was retrieved.