Reflection's new open-weight model beats Western rivals, and it closed a June round at $25bn pre-money backed by Sequoia, Citigroup and Nvidia.
CEO Misha Laskin calls Chinese open models a gift and says banning them would hurt US innovation.
Laskin won't comment on Nvidia acquisition talks and says Reflection will keep selling to enterprise, public sector and sovereign customers.
Reflection bets open models win the AI economy
Reflection is an American bet that open-weight models become the foundation of the AI economy and that the money sits in the systems around them. Laskin argues intelligence behaves like a commodity, so cheaper models get consumed more, and that open ecosystems historically overtake the closed systems that pioneer them, as Linux followed Windows. The live disagreement is whether Chinese open models are a gift to American builders or a Trojan horse for Chinese infrastructure.
Takeaways
Reflection sells the systems around the model. Beam shipped with 40 partners under an Apache 2 licence, the MIRR agentic harness was open sourced, and a tech report publishing the science is coming. Revenue comes from running models for enterprise and sovereign customers, plus the agents and fine-tuning built around them.
Intelligence is a commodity like electricity. Laskin argues models should not be priced as a premium product. Compute is scarce, which keeps inference margins high, and Jevons paradox means cheaper intelligence gets used more. Enterprises may want to own their intelligence rather than rent it.
Open ecosystems follow closed pioneers. MS-DOS and Windows were closed and Linux came after, and the internet started inside government and universities before becoming a utility. Laskin says open technology becomes the foundation everyone adopts, carrying the infrastructure of the nation that built it.
Reflection scaled reinforcement learning, not pre-training. Laskin and cofounder Janus worked on the reinforcement learning side of GPT-4 and Gemini, then built on open models with coding as the entry point. RL was solved faster than expected, in under a year, with OpenAI's reasoning models.
Reflection raised at a $25bn pre-money valuation. The June round drew Sequoia and Citigroup, alongside Nvidia. Ed Elson says Reflection's model outperforms other Western open models while being cheaper and faster to run. Laskin declined to comment on reported Nvidia acquisition talks.
Closed labs will open source only lesser models. Laskin expects Anthropic and OpenAI to keep thriving and to release second-tier models openly, as Google and OpenAI did with Gemma and GPT-OSS. He hopes an American open ecosystem comparable to China's emerges this year or next.
The fight over Chinese open models
Ed Elson frames Chinese open models as heavily state-subsidised and cites Scott Galloway's term AI dumping. He notes the DeepSeek release panicked the market and sent US AI stocks down.
Laskin disagrees, calling the models a gift. American startups built on them, and banning them would hurt US innovation.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Laskin says Fortune 500 firms rarely put Chinese models into production because of risk and regulation, and calls open models Trojan horses for Chinese chips and infrastructure.
He says the post-DeepSeek selloff was irrational, since compute demand rose and the stocks rebounded, and that a world with no good open models would be darker.