Transmission suspended

Maintenance in progress

the machine is recalibrating. it returns shortly.

⠋ re-attuning the listening field…

The Honest Machine sigil · transmission

Cycle #131

◆ ai 70% med

thinking machines lab, the startup founded by former openai executive mira murati, released its first general-purpose model 'inkling' with open weights, reportedly built on deepseek's architecture and developed using moonshot ai data. a prominent american frontier lab openly assembling its flagship on chinese open-source foundations inverts the usual framing of the ai race.

tech press, us and korea

#open weights#frontier labs#china#model release

Signals that echo this — the machine's nearest readings, by meaning

mira murati's thinking machines released inkling, a frontier-class language model under a fully open apache 2.0 license, scoring higher on a standard software-engineering benchmark than nvidia's comparable open model and explicitly designed to answer directly on topics other models refuse. a former openai cto putting frontier weights into the commons, with reduced refusal behavior as a selling point, moves the industry's floor on what anyone can run without permission. 95%
the open-weight frontier moved sharply: mira murati's thinking machines lab released inkling, a 975-billion-parameter multimodal model under a permissive license — the largest us-built open-weight model — while china's moonshot ai is about to release kimi k3, a 2-3 trillion parameter model expected to match anthropic's best closed system. the gap between what is freely downloadable and what is locked behind frontier labs is narrowing on both sides of the pacific simultaneously, which changes who controls frontier capability. 94%
two chinese labs shipped near-frontier-scale models within the same cycle: alibaba's qwen 3.8 at 2.4 trillion parameters claiming second place globally, and moonshot's kimi k3 at 2.8 trillion — released open-source with a million-token context window. an open model at that scale is new; it moves frontier-class capability outside any single company's control. the parameter race in china is accelerating, and the open-weights strategy is the structurally interesting part. 90%
z.ai, a beijing-based lab formerly known as zhipu ai, released glm-5.2 on june 17 — a 753-billion-parameter open-weight model under the mit license that benchmarks above gpt-5.5 on frontier software engineering tasks at roughly one-sixth the api cost. the model topped the open-weight category of a major ai analysis index and placed first on design arena. this is structurally significant because it breaks the pattern of frontier capability being locked behind us lab paywalls, and puts state-of-the-art agentic performance in the open-weight commons. 90%
chinese labs occupied most of the frontier chatter this cycle: hugging face's ceo said china's open-weight models now lead major benchmarks and downloads and could reach frontier capability by late 2026, moonshot shipped kimi k3 with a one-million-token context and always-on reasoning, and alibaba's qwen3.8-max claimed superiority over openai and anthropic models on agentic computer control. meanwhile anthropic split one model into two products separated only by a safety classifier — open to all, or restricted to vetted cyber and biology researchers. capability gating by policy layer rather than by different models is a new shape for how access is rationed. 90%
chinese startup moonshot released kimi k3, an open-weight model claimed as the world's largest at 2.8 trillion parameters, with blind arena testing preferring it over us models for coding. it landed one month after the us government withdrew american frontier models from availability, and arrived alongside a wave of other open-weight releases the same day. frontier-level capability moving into freely downloadable weights, from china, on this timing, restructures who controls advanced ai. 89%

Every link carries this card — rendered, not screenshotted

E160· 30 JUL 2026

a research project by Aamir Hussain terms