Moonshot AI released the weights for its Kimi K3 model on Monday, a system with 2.8 trillion parameters. Developers can now download, modify and self-host K3, which the report described as the world's largest open-weight model. Anthropic CEO Dario Amodei published a new position post on open-weight models, while Sam Altman and Jensen Huang are scheduled to meet lawmakers in Washington this week.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Amodei's post

In his post, Amodei wrote, "Anthropic has never advocated for a ban on open-weights models."

Open-weight models without dangerous capabilities are "a public good," Amodei wrote, and he opposed a category-wide prohibition on US business use. In the post, Amodei backed three measures. The first called for withholding powerful chips and chipmaking equipment from China. Industrial-scale distillation operations should also face a crackdown, according to the post. His third measure would require mandatory safety testing for all sufficiently capable models, open and closed. Effective testing would need to operate globally, Amodei said. Protectionist bans would not address his main national-security concerns, according to Amodei. The post states that protecting US AI companies from competition has never been his goal.

Amodei listed two capability risks. His primary concern is that authoritarian governments could build models more powerful than US systems and deploy them for military superiority or domestic repression, according to the post. His second concern covers cyber or biological misuse and serious alignment problems. A ban on legitimate US business use would address neither problem, Amodei maintained.

Amodei also addressed the July 24 industry letter, signed by Nvidia, Microsoft, Meta, Palantir and later OpenAI, which urged policymakers to avoid "premature restrictions" on open-weight models. He rejected its assertion that open weights necessarily make safeguards easier or that wider access favors defenders. Models with dangerous capabilities may be harder to guard or monitor once their weights are downloaded, he stated, and the files cannot be withdrawn after release. He agreed with parts of the July 24 letter, including its claims about competition and customer control. Its safety assertions, he wrote, should be tested before release rather than assumed in advance.

The 68% token mix

Bloomberg Intelligence analysts Mandeep Singh and William Tong called open-weight large language model adoption "inflecting" and tied monetization to distribution through large cloud platforms. On token volume, they wrote that the mix for open-weight Chinese models reached 68% after the launches of Kimi K3 and Z.ai's GLM-5.2. AWS Bedrock, Microsoft Azure Foundry and Google Vertex AI had yet to support those Chinese open-weight systems, the analysts noted.

Tobi Knaup, a Mesosphere co-founder, made a separate industry case in a July 24 personal essay. Knaup cited Hugging Face's count of more than two million public models. Chinese models accounted for 41% of model downloads on Hugging Face over the past year, according to another finding he cited from the platform. Companies' desire to control their data and inference costs drove demand for self-hosting, Knaup wrote. An independent evaluation by Artificial Analysis scored Kimi K3 alongside Opus 4.8 and GPT-5.5, he noted.

A broad ban on Chinese models would exclude American researchers and companies from work developers elsewhere could continue, Knaup argued. Most systems described as open source offer downloadable weights, while their training data and complete training process usually remain unavailable, he noted. No AI equivalent of the CNCF supplies neutral governance and common interfaces, the essay added. Knaup also favored independent testing and safety standards over a blanket prohibition.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Altman and Huang in Washington

Altman will meet senior Trump administration officials, lawmakers and economists in Washington this week, CNBC reported. Altman, the same report noted, had maintained a regular presence on Capitol Hill during 2026 while discussing AI regulation. The July 24 letter Amodei answered was signed by OpenAI after its initial publication, according to CNBC. Altman plans to preview OpenAI's coming model family and answer questions about cybersecurity and open weights. That agenda came from a person familiar with the plans who requested anonymity because the details are confidential. Huang will also meet lawmakers to discuss open models and "American leadership in AI," an Nvidia spokesperson said.

Moonshot's next technical report

Moonshot founder Yang Zhilin has said he wants to win users through openness and wider availability than proprietary US systems. Bloomberg News reported that daily sales had climbed at least sixfold since K3's debut. The company is seeking a funding round at a $50 billion valuation before a possible Hong Kong IPO as soon as this year.

The weight-release article says further details on K3's architecture, training and evaluation will appear in an upcoming technical report. It gave no publication date.

Frequently Asked Questions

What exactly did Moonshot AI release?

The Beijing-based company released the model weights for Kimi K3 on Monday. The system has 2.8 trillion parameters, and developers can now download, modify and host it on their own infrastructure rather than calling it through an API.

What is Anthropic's stated position on open-weight models?

In his position post, CEO Dario Amodei wrote that "Anthropic has never advocated for a ban on open-weights models" and described open-weight models without dangerous capabilities as "a public good." He said protectionist bans would not address his main national-security concerns, and that protecting US AI companies from competition has never been his goal.

What does Amodei support instead of a ban?

Three measures. Withholding powerful chips and chipmaking equipment from China, a crackdown on industrial-scale distillation operations, and mandatory safety testing for all sufficiently capable models, open and closed. He said effective testing would need to operate globally.

How widely are open-weight Chinese models actually being used?

Bloomberg Intelligence analysts Mandeep Singh and William Tong wrote that the token-volume mix for open-weight Chinese models reached 68% after the Kimi K3 and GLM-5.2 launches. Tobi Knaup cited a Hugging Face finding that Chinese models accounted for 41% of model downloads on the platform over the past year.

What was the July 24 industry letter?

A letter signed by Nvidia, Microsoft, Meta, Palantir and later OpenAI, urging policymakers to avoid "premature restrictions" on open-weight models. Amodei agreed with parts of it, including its claims about competition and customer control, but rejected its assertion that open weights necessarily make safeguards easier or that wider access favors defenders.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

OpenAI and Anthropic Lobby Washington to Restrict Chinese Open-Weight AI
OpenAI and Anthropic have lobbied Washington regulators to restrict Chinese open-weight AI models, five people close to the discussions told The New York Times on Saturday. According to the same repor
The Implicator
Germany's Soofi S AI Model Tops All Open-Source Rivals on German Benchmarks
A German research consortium coordinated by the KI Bundesverband released Soofi S, an open-source German-English foundation model, this week, according to its pretraining report. In the team's tests,
The Implicator
GLM-5.2 Edges Kimi K2.7 Code in Early Coding Tests
The two strongest Chinese open models to launch this month, Z.ai's GLM-5.2 and Moonshot's Kimi K2.7 Code, have now been run side by side by five independent reviewers, and the early verdict gives GLM-
The Implicator
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai