for two years, the constraint was simple: not enough gpu supply. now the workload itself is changing, and with it the bottleneck. ai systems are moving from generating answers to
Nick Trenkler
5 min read
openclaw claims claude cli is “allowed again” for agent-style usage (not confirmed publicly by anthropic) at the same time, anthropic just deepened its deal with amazon – $5b invested now, with
Nick Trenkler
1 min read
kimi just dropped k2.6 – and did it the new way: model + agent, shipped together this is already the trend: • gpt + openclaw. openai aligning openclaw to its models, turning it
Addy Crezee
1 min read
their updated policy claims that: • non-coding usage may lead to restricted benefits • violations can trigger high-intensity throttling, suspension, or permanent ban • three or more violations result in a permanent account
Addy Crezee
1 min read
1/ fact. since 2023, deepseek declined offers from china’s top vcs and tech giants and stayed fully funded by its hedge fund. now it’s opening up to external
Nick Trenkler
1 min read
did openai secretly upgrade gpt pro? x thinks so. for the last 24 hours, x has been filling up with the same kind of claim: gpt-5.4 pro suddenly feels
Nick Trenkler
2 min read
x api is now accessible via openclaw – and this is bigger than it looks x api includes: • posts (create, publish, edit, and delete posts; read timelines, threads, replies, and quote
Nick Trenkler
1 min read
xai is giving tens of thousands of gpus to cursor to train composer 2.5 – external workloads, not just internal models and that raises the real question: what does xai
Addy Crezee
1 min read
if you want to run strong models without living inside the cloud bill, april 2026 looks pretty fun: 1. qwen 3.6-35b (@Alibaba_Qwen) – 73.4 swe-bench on a 24gb
Julia Daimio
1 min read
peter gostev reported that opus 4.7 performs worse than the opus 4.6 family on bullshitbench – a benchmark that measures whether models can detect nonsense and refuse to answer
Nick Trenkler
3 min read
opus 4.7 launched. thehype reviewed early x feedback. here’s what users are actually seeing: 1/ consumes ~50% more tokens than opus 4.6 and ~2x vs gemini efficiency
Nick Trenkler
2 min read
сlaude opus 4.7 is live! numbers that matter (4.7 vs. 4.6): - agentic coding: 87.6 vs 80.8% - terminal coding: 69.4 vs 65.4%
Addy Crezee
1 min read
minimax is making its move they’re teaming up with @nousresearch to push m2.7 as the default brain for hermes agent – and going straight for mass adoption with maxhermes
Nick Trenkler
1 min read
the original opus 4 shipped eleven months ago. it already feels obsolete. opus 4.7 is +26 points on terminal coding, +15 on agentic coding, +11 on reasoning. the models
Nick Trenkler
3 min read
3D world models are having a moment but only a couple let you just go and create the rest need time and developer skills we ranked them by accessibility. check
Addy Crezee
1 min read