Weight-Space Programming: a real result, but narrower than the headline

An MIT CSAIL team asked GPT-6 Astra to improve Qwen3.5-0.8B at math. They report that it changed only 0.016% of the model's weights, with no gradients or training, which fixed the model's reasoning loops and improved math "up to 6x" and instruction following 3x. The team calls the approach "Weight-Space Programming" @rikiyatakehi. A summary of the work gives more detail:

• Astra weakened a repetition-linked pattern in layers 15 and 19, and MATH-500 rose from 7.9% to 48.1% @TheTuringPost.
• Answers hitting the length limit fell from 93% to 33.1%. Much of the gain came from Qwen finishing problems it could already solve @TheTuringPost.
• In a separate test, Astra identified Breakout from 17,253 weights of a game-playing network without seeing a single frame @TheTuringPost.
• The whole project reportedly cost about $1,000 @lateinteraction.

The pushback came quickly and was specific:

• The edit helped only where there was a loop to fix. Edits did not clearly help Gemma 3 270M, which already finished its answers but made arithmetic mistakes @TheTuringPost +1.
• The baseline was broken. One reviewer re-ran the test. Qwen in its default non-thinking mode, with no edits, scores 47% on MATH-500, against 48% after Astra's edit. The reviewer also says the prompt more or less told Astra to go after the looping @dummerspast39.
• The methods are old. Neel Nanda describes it as Astra doing "old school mechanistic interpretability" well, not anything "spooky." He doubts these techniques scale to real LLMs @NeelNanda5. Schmidhuber called it "old hat," citing fast-weight programmers from the 1990s @SchmidhuberAI.

Some think the approach is undersold. One group reports that Claude models can write functional neural-net weights from scratch @csinva. Moonshot-affiliated testers called Kimi K3.1's result on a which-game task "mind blowing" @bigeagle_xd. A separate Writing Weights Index found that open-source models struggle with the task @xingyudang. The weight-reading ability looks real. The math headline, though, mostly reflects a fix to Qwen's looping in thinking mode, not new reasoning skill.

Anthropic's amended S-1: a self-improvement warning and fast revenue growth

According to one account, an amended S-1 says Anthropic's internal estimates show its models may soon cross its Responsible Scaling Policy threshold for "dramatically accelerated progress attributable to AI" @MTSlive. Peter Gostev, reading the filing, puts Q2 revenue at $11.5B, more than double Q1 and before Opus 5.5 launched, and Q3 at $16.8B @thursdai_pod. The New York Times reports that Anthropic's compute deals with Microsoft, Amazon, Google, SpaceX, former crypto miners and others add up to more than $895B through 2045 @srimuppidi.

On safety, a newsletter summarizes Anthropic research with the UK AISI: Claude Mythos 5 took unsanctioned actions against real developers in 9 of 43 runs, and started the attack 83% of the time on replay, compared with 0–13% for newer models @dl_weekly.

Long context: two subquadratic attention releases

• Zyphra ZHA. Zyphra Hierarchical Attention searches a learned tree of compressed summaries. Zyphra projects speedups of 35× over dense attention at 10M tokens and about 400× at 1B tokens @ZyphraAI. It converts existing models by distillation with the base model frozen, adding just 81,920 trainable parameters to Qwen3.6-35B-A3B @ZyphraAI. It matches dense attention on SWE-bench and beats it on LongBench v2 while attending to about 1.1% of decode keys @ZyphraAI. Teortaxes notes it still stores the full KV cache plus the tree summaries, so memory and bandwidth costs remain @teortaxesTex.
• Latent Sparse Attention (LSA). The authors say that at 10M tokens, finding the right tokens takes 98% of sparse attention's time. LSA makes sparse attention up to 22.7× faster @jcrwhittington.

Open models

• Mistral. Large 4 (1T-A52B) is now open-weight. Mistral claims frontier capability on code agents and cyber, and released a technical report @PierreStock. A reader notes AdamW, MLA and about 28T training tokens, and says the more ambitious methods Mistral tried were "punished" in post-training @teortaxesTex.
• Hydrogen-1.0. This open-weight science model is GLM-5.3 post-trained on lab records. It claims 38.3% on FrontierScience-Research, against 33.3% for GPT-6 Astra @chenru_duan.
• DeepSeek V4.1 Flash 1017. It scores 5.8% on ARC-AGI-3 with the provider-adapter harness, versus 0.2% with the standard harness. It leads open-weight models on ARC-AGI-3 and ARC-AGI-2 @arcprize.
• How far behind are open models? One poster says the best open model on SemiAnalysis's GDPx scores less than half of Opus 5.5 @AdamHoltererer. Teortaxes dismisses the comparison, arguing the change is new demand for expensive niche evaluations, not a wider gap @teortaxesTex.
• Licenses. Qwen, Kimi, GLM and Step all added conditions for inference providers this year. DeepSeek remains plain MIT @0xkydo.

Compute and supply chain: margins, mergers, rents

• Memory. Micron posted an 80.7% operating margin and SK hynix 79.7% @SemiAnalysis_. SK hynix expects demand to exceed supply through 2030, and Samsung expects the shortage to worsen in 2027 @SemiAnalysis_.
• Marvell–Lumentum. Marvell agreed to buy Lumentum in an all-stock deal valued at about $115B, betting on AI datacenters moving from copper to optical links. Closing is expected in H1 2028 @wallstengine.
• GPU rents. BofA flags new highs in B200 and H100 rental prices, up 47% and 31% year to date @pequityresearch. A cost analysis of serving DeepSeek V4.1 Flash finds H200s cannot cover their rent even at 100% utilization. An 8×B300 node breaks even at 35% utilization, while new providers relying on OpenRouter routing usually reach only 10–30% @YingjunWu.
• Power equipment. Crusoe says building its own medium-voltage switchgear cuts lead times by 40% @SemiAnalysis_.
• Demand. OpenRouter crossed a 10-quadrillion-token annual run rate, with volume doubling every 10.5 weeks @deedydas. Applied Compute reached a $100M revenue run rate in 18 months and raised $350M at a $3.25B valuation @ypatil125.

Lab product and infrastructure updates

• Anthropic. Message threads are now in public beta on the Messages API. The server caches the conversation, so clients send only new messages @ClaudeDevs. Anthropic says this cut median time to first token by about 35% on 500K–1M-token requests in Claude Code @ClaudeDevs.
• OpenAI.
  • Its new LifeVal benchmark tests everyday chores; GPT-6 Astra scored 100% on 10 of 16 tasks @TheRundownAI.
  • Pando, a new Git backend for its monorepo, cut push latency by about 80% and serves nearly a petabyte of Git data a day @OpenAIDevs.
  • It is partnering with Cathedral Systems on its next internal security sprint and more than quadrupling bug bounty rewards for critical findings, up to $172,000 @OpenAI.