Article · 2026-07-20 · 12 moments

Kimi K3: The open-weights escalation

The global implications on the AI ecosystem. ✦ AI generated

Fact

Alibaba's surprise announcement of a 2.4 trillion parameter open-weight Qwen 3.8 model, coming right after Kimi K3's release, shows Chinese labs are leaning further into open-weight releases rather than just maintaining the status quo.

Lambert points to Alibaba breaking from its API-only tradition for large models by announcing an open-weight 2.4T Qwen model days after Kimi K3, as further evidence of a broader Chinese pivot toward open release.

transcript

Nathan Lambert: Alibaba announced that a 2.4 trillion parameter Qwen 3.8 model is coming soon with open-weights. Historically, Alibaba has kept their largest models as API-only offerings via their cloud business, so this is another big vibe shift opening the doors to the next chapter of the open model economy.

extends · 1provides context · 1

Claim◆

Banning or heavily restricting open-weight models in the U.S. would create a dangerous asymmetry, where American closed models carry cybersecurity guardrails but global adversaries still have unrestricted access to capable Chinese open-weight models to probe U.S. defenses.

Lambert warns that U.S. policy moves to restrict open-weight models would leave guardrailed American models at a disadvantage while adversaries retain access to strong Chinese open-weight alternatives, arguing bans harm both markets and security.

transcript

Nathan Lambert: This would leave the U.S. in a very asymmetric state where the best models in the U.S. have guardrails on cybersecurity tasks, but global actors have access to great Chinese open-weight models to probe our defenses. This is one of many examples where banning open-weight models is not only harms the free markets of AI but also makes the ecosystem less safe in the short-term.

extends · 1rebuts · 3supports · 2

Claim

Chinese AI labs are far more capital efficient than their American counterparts, and since intelligence scales with effective capital, this efficiency may be the single greatest strength of China's AI industry.

Lambert argues that despite having far less capital than U.S. labs, Chinese companies like Moonshot are matching or beating them on benchmarks, pointing to a decisive efficiency advantage in training.

transcript

Nathan Lambert: It is becoming clear that the Chinese labs are far more capital efficient. In a world where scaling laws dictate that intelligence is proportional to effective capital – which buys compute, data, & talent – that may be the greatest strength your AI industry could ever have.

extends · 1

Claim

Chinese AI labs are far more capital-efficient than their American counterparts, converting proportionally less capital into competitive model performance.

Lambert argues that despite having orders of magnitude less capital raised than US labs, Chinese labs like Moonshot achieve outsized results, suggesting a structural efficiency advantage.

transcript

explains mechanism · 1supports · 1

Prediction

Heavy-handed regulation of open-weight models would not prevent their risks, only briefly delay the inevitable spread of capabilities while lulling people into false complacency.

Lambert concludes that since open models cannot be effectively banned globally, harsh regulation mainly creates false comfort rather than actually stopping capability diffusion.

transcript

Nathan Lambert: If we regulate open-weight models heavy-handedly, I suspect much of the world will be lulled into thinking we no longer need to act. All we would've done is slightly delayed the inevitable — open models will continue to cross all the key capability thresholds eventually and regardless of legality.

Claim◆

If adversarial distillation from closed U.S. frontier models contributed to Kimi K3 at all, it was to a relatively small degree — Chinese labs like Moonshot AI are genuinely solving the same hard problems as OpenAI and Anthropic, not just copying their outputs.

Nathan Lambert argues Kimi K3's strength shows Chinese labs can build frontier models through genuine execution, not mainly by distilling U.S. models, undercutting the 'IP theft' narrative around Chinese AI.

transcript

Nathan Lambert: It should be clear looking at this model that if adversarial distillation from the closed frontier models in the U.S. contributed, it is at most to a relatively small degree. AI observers who followed the distillation panic and came away with the wrong conclusion that Chinese AI labs are only producing good models due to IP theft are in for an awakening

supports · 1

Claim◆

Kimi K3's quality shows Chinese labs like Moonshot are genuinely skilled model builders, not merely distilling stolen output from closed American frontier models.

Lambert argues the strength of Kimi K3 disproves the 'distillation panic' narrative that Chinese labs only succeed via IP theft from US models.

transcript

Nathan Lambert: It should be clear looking at this model that if adversarial distillation from the closed frontier models in the U.S. contributed, it is at most to a relatively small degree... that Chinese companies are extremely good at building models in the same way the leading American companies are.

supports · 1

Claim

Open-weight models are inherently decelerationist for the AI buildout, which is ironic given that self-described accelerationists tend to champion them.

Dean Ball's widely-discussed claim, quoted and endorsed by Lambert, that open-weight models undercut frontier labs' margins and thus slow capital reinvestment into next-generation AI.

transcript

Dean Ball: Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models.

provides context · 1

Claim

Open-weight models are inherently decelerationist for the AI race, which is why it's surprising that self-described accelerationists are so enthusiastic about them.

Dean Ball, quoted by Lambert, argues open-weight models slow the frontier by cutting into the margins and capital that closed labs need to fund ever-larger training runs.

transcript

Dean Ball: Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models.

provides context · 1

Claim◆

The performance gap—both open-to-closed and American-to-Chinese—has shrunk from a debated 6-9 months down to something like 3-5 months.

Nathan Lambert argues Kimi K3's release shows the open/closed and China/US capability gaps have compressed to roughly 3-5 months.

transcript

Nathan Lambert: The key fact is that either the open-to-closed or American-to-Chinese model performance gap has been reduced from the debated 6-9 months to something shorter, say 3-5 months.

rebuts · 1

Fact

Xi Jinping's keynote at the World AI Conference directly committed China's AI ecosystem to open-source release and global diffusion, the first time senior Chinese leadership has publicly staked out a national position on open-source AI — and it landed the same week as the strongest open-weight model release to date.

Lambert notes that Xi Jinping's WAIC address was the first senior-leader comment on open-source AI policy in China, coinciding with the Kimi K3 release, signaling low perceived risk from open-weight frontier models.

transcript

Nathan Lambert: This changed this week too, as Xi Jinping gave a keynote address at the World AI Conference (WAIC), and very directly committed the future of China's AI ecosystem to open-source and global diffusion. This commitment to the status quo, the same week as the announcement of the strongest open-weight model to date, is a clear mark in the early history of modern AI.

Claim

Xi Jinping's WAIC keynote committing China's AI ecosystem to open-source and global diffusion, timed alongside Kimi K3's release, reveals that China's government does not currently see meaningful risk (e.g. cybersecurity, bio) in frontier open-weight models.

Lambert reads Xi's WAIC speech, coinciding with the strongest-ever open model, as an implicit signal that China's leadership assesses low risk from current frontier capabilities.

transcript

Nathan Lambert: Xi Jinping gave a keynote address at the World AI Conference (WAIC), and very directly committed the future of China's AI ecosystem to open-source and global diffusion... The simple explanation is that they do not find current frontier models to have meaningful risk.

provides context · 1rebuts · 2

Highlight slides

AI Capability Gap Has Shrunk✦ from: The performance gap—both open-to-closed and American-to-Chinese—has shrunk from a debated 6-9 months down to something like 3-5 months.Kimi K3: Genuine Engineering, Not Copying✦ from: If adversarial distillation from closed U.S. frontier models contributed to Kimi K3 at all, it was to a relatively small degree — Chinese labs like Moonshot AI are genuinely solving the same hard problems as OpenAI and Anthropic, not just copying their outputs.Banning Open-Weight Models Creates a Security Gap✦ from: Banning or heavily restricting open-weight models in the U.S. would create a dangerous asymmetry, where American closed models carry cybersecurity guardrails but global adversaries still have unrestricted access to capable Chinese open-weight models to probe U.S. defenses.Kimi K3 Undercuts the 'Distillation Panic' Narrative✦ from: Kimi K3's quality shows Chinese labs like Moonshot are genuinely skilled model builders, not merely distilling stolen output from closed American frontier models.Nathan Lambert's Core Verdict✦ from: Kimi K3's quality shows Chinese labs like Moonshot are genuinely skilled model builders, not merely distilling stolen output from closed American frontier models.Gap: Before vs Now✦ from: The performance gap—both open-to-closed and American-to-Chinese—has shrunk from a debated 6-9 months down to something like 3-5 months.Why Bans Backfire✦ from: Banning or heavily restricting open-weight models in the U.S. would create a dangerous asymmetry, where American closed models carry cybersecurity guardrails but global adversaries still have unrestricted access to capable Chinese open-weight models to probe U.S. defenses.A Reckoning for the Distillation Panic✦ from: If adversarial distillation from closed U.S. frontier models contributed to Kimi K3 at all, it was to a relatively small degree — Chinese labs like Moonshot AI are genuinely solving the same hard problems as OpenAI and Anthropic, not just copying their outputs.

Related episodes

Video · a16z Podcast

Ben Horowitz on the Global Race for Tech, Power, and Influence

→

Audio · All-In Podcast

OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze

→

Video · All-In Podcast

OpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts

→