Inception Labs
(San Francisco, CA)
Diffusion-LLM specialists backed by Eric Schmidt, the Collisons, Karpathy, Nat Friedman, and Andrew Ng; Mercury 3 generates tokens in parallel at 1,000+ tok/s to outpace speed-optimized frontier models.
What is the latest AI model from Inception Labs?
Inception Labs's primary flagship model is Mercury 3 (Undisclosed (40x+ larger than Mercury 1), 1,000,000 tokens context). First diffusion LLM in the registry: Inception's Mercury 3 (Oct 19, 2026) generates tokens in parallel rather than left-to-right - 1,068 tokens/sec, sub-300ms time-to-first-token, outperforming GPT-6 Luna and Claude Haiku 5.5 across benchmarks while running 4-8x faster; 1M context; launch price $0.10/$0.35 per 1M tokens (50% off the standard $0.20/$0.70), available via API, OpenRouter, Baseten, and Vercel. How we define “latest”