A lot of hype around OpenAI's Astra model here on my timeline today. Apparently, this goes back to a new article from The Information, which said Astra is a "recurrent depth or looped transformer". It's always interesting to read about new or different approaches (including rumors about what the closed labs may be up to), but let's debunk this a bit. About 2 months ago, I shared the architecture details of Nanbeige, for example, where "Nanbeige4.2-3B is pretrained from scratch on 28T tokens with a Looped Transformer that reuses the layer stack to increase capacity without adding parameters." Yes, that's it. The looped transformer idea is just reusing layers in the transformer block. In the case of Nanbeige, the main idea is to reuse the same 22-layer stack (=transformer block) twice instead of once. So, effectively it extends the 22-layer architecture to 44 layers, but without duplicating the weights. In simple terms, this roughly doubles the size of the model (if we ignore the embedding and output layers for a second). But instead of requiring 2x the storage and RAM to host this model, it stays at the same size since we reuse the components. However, it's almost 2x as expensive in terms of compute, because we run the embedded text through almost 2x as many layers. Why? In the Nanbeige 4.2 technical report, the researchers found that two passes gave the best trade-off and retained about 75% of the token efficiency of a standard architecture. (More passes gave barely any gains but made the training much slower and much more expensive.) While, as far as I know, Nanbeige 4.2 is the first notable open-weight model that adopted this approach, the idea goes back to the NeurIPS paper "Mixture-of-recursions: Learning dynamic recursive depths for adaptive token-level computation". Actually, this paper proposes a mechanism that is a bit more sophisticated by adding a learned router that determines whether each token receives one, two, or more passes. So, easy tokens can exit early while harder tokens receive additional computation. In sum, Astra may be a really good model, but this shouldn't be about this "looped transformer aspect," which is just a tiny architectural tweak. Also, the statement "the new technique works in a way that obscures some or all of the AI's reasoning, otherwise known as 'chain-of-thought'" is not necessarily true with respect to the looped transformer method. It's possible that The Information journalist refers to some other technique or misunderstood the looped transformer method. Reusing layers does not by itself suppress visible chain of thought. It adds computation in hidden states before the next token is emitted, just as ordinary transformer layers do. But based on the information we have, the only plausible interpretation here is that if a model uses more of these recurrent passes, it may need to generate fewer intermediate reasoning tokens. So then more of its computation happens in latent activations that cannot be read as text. But we would get the same effect if we were scaling up the model size, like GPT 5.6 Luna -> GPT 5.6 Sol.

DeepSeek Harness 发布短短两天，GitHub Star 已经突破 10 万。 它已经远远超过存在时间更长的 Qwen Code，也超过了非常受欢迎的 Pi，甚至已经超过 OpenAI Codex。 目前前面只剩 OpenCode。 更夸张的是，这些项目大多已经积累了几个月甚至更长时间，而 DeepSeek Harness 只用了两天。 以目前这个势头，超过 OpenCode，恐怕也是指日可待。

Two months ago I was testing Grok @imagine to make the Battle of Red Cliffs (赤壁之戰). Now in Grok @bot you can spin up a director, screenwriter, music producer, and cinematographer to create a cinematic movie trailer in one shot. Attached the only prompt I gave to Grok @bot and the rest was autonomous.

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12. We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

Kimi K3 exposed the real bottleneck in open AI: open weights do not mean easy inference. Moonshot paused new subscriptions as demand overwhelmed capacity, and recommends 64+ accelerator supernodes for K3. This also sheds light on what the model-serving ecosystem may look like once K3 releases its weights. It turns out we already have a thriving industry: @FireworksAI_HQ , @baseten and @togethercompute carry $38.8B in combined valuations, with revenue, bookings and token volumes growing rapidly.

The agent cyber war has begun, and GLM-5.2 helped lead the counterattack. Hugging Face recently disclosed a security breach in which an autonomous AI agent swarm executed more than 17,000 actions, breached its infrastructure, harvested credentials, and moved laterally across internal clusters. Hugging Face first tried using commercial frontier models, presumably Mythos or Fable 5, to investigate. But their safety guardrails blocked the actual exploit payloads and attack commands. So what did they use for the counterattack? GLM-5.2. They ran the open-weight model on their own infrastructure to reconstruct the attack, trace compromised credentials, and separate real damage from decoys. An AI agent attacked. Another AI agent, powered by a local open model, helped fight back. @Zai_org

Huawei's Atlas 950 SuperPoD debuts at WAIC 2026. The Atlas 950 SuperPoD is its flagship AI compute system for model training and inference. Built from up to 8,192 Huawei Ascend NPUs and linked by Huawei’s UnifiedBus interconnect, it is designed to deliver ultra-high bandwidth, ultra-low latency, and unified memory addressing.