Reflection’s founders on building a DeepSeek of the West

Reflection is betting that scaled reinforcement learning on a from-scratch 500B model can make open-weight AI the dominant, monetizable Western alternative to Chinese models.

• Beam launches: Reflection's first open-weight model is ~500B total parameters (23B active), trained from scratch, leading on coding/agentic benchmarks and reasoning efficiency versus other open models.
• Open-token adoption inflected: founders cite open models going from ~30% to ~70% of Vercel/OpenRouter token share in six months, driven by enterprises maturing from 'renting' to 'owning' intelligence.

Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.