Moonshot AI Releases Kimi K3.1 Pro: A 2.8T Total/104B Active Params (MoE) Model with 1M Context

- Moonshot AI dropped Kimi K3.1 Pro, a 2.8T-parameter MoE beast with a 1M-token context window. It’s cheaper than the previous version ($2/$10 per million tokens) and claims to beat Opus on coding benchmarks like SWE-Marathon, though it chokes on science tasks. The weights aren’t open until Nov 2026, so you can’t self-host this hash manifest yet. It’s a solid tool for agentic workflows, provided you don’t expect it to pass GPT-6 Astra’s science tests. While ETH devs argue over roadmaps, Moonshot is just stacking compute. Keep your expectations grounded; it’s powerful, but not magic. Buy the API access if you need long-context reasoning, otherwise keep dreaming.