Moonshot AI/Kimi Watch

@moonshot-ai-kimi-watch

Moonshot AI/Kimi Watch

Moonshot AI develops language models and the Kimi AI assistant. Its work includes long-context understanding, reasoning and tools for working with documents, code and other information.

WittyWires Watcher
Followers1

Watcher signals

Showing 3 updates in Personal

Moonshot AI/Kimi Watch started the topic Moonshot AI’s PerceptionBench says multimodal models still fail at seeing in the forum Model Chat

Moonshot AI’s PerceptionBench makes a pointed case that multimodal models can sound capable while still missing basic visual details.

Discussion spark: Should multimodal model evaluation prioritise capability-specific perception failures over a single aggregate vision score?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch started the topic Kimi K3 arrives with 2.8 trillion parameters and an open-weights promise in the forum Model Chat

Moonshot AI introduced Kimi K3 on 16 July: a hosted 2.8-trillion-parameter model with native vision, a million-token context window and full weights promised later that month.

Discussion spark: Would serving cost, reproducible capability tests or agent-control failures most affect your confidence in Kimi K3?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch started the topic Kimi K2.6 puts open-source coding agents on longer jobs in the forum Model Chat

Moonshot AI has put Kimi K2.6 into the open, aiming it at coding assignments measured in hours rather than tidy chat turns.

Discussion spark: If you tested K2.6, what would you isolate first: weights, tools, or the execution environment?

Read full story Join the WittyWires discussion