• Moonshot's most capable open model so far, in Moonshot's words. It is built on Kimi K3, with much more training on top.
• It reads a lot at once. About 1 million tokens (word pieces), Moonshot says. That is a very long document, or all the code of a large project.
• Open-weight. Moonshot says the full model files will be released by 4 November 2026, so anyone will be able to download and run it. They are not released yet.
• Where it improved. Moonshot reports gains in coding, 3D design, finance research and agent work: an AI that carries out multi-step tasks on its own.
• A faster version. Kimi K3.1 Pro Highspeed runs two to three times as fast, Moonshot says.
• Where to use it. Today, in the Kimi website and apps, Kimi's coding tool, and through Moonshot's paid developer access.

These are Moonshot's own numbers, from its own tests with its own settings. Nobody outside Moonshot has checked them yet.

Price for developers: $2 for every million tokens it reads, and $10 for every million it writes. That is cheaper than Kimi K3, at $3 and $15. This is Moonshot's price list, as reported by others.

• No outside tests yet. Nobody outside Moonshot has published their own test of the model yet. The numbers above are Moonshot's own.
• Near Opus, not past it. One writer called it the closest an open model has come to Claude Opus 5.5, priced like a mid-range model. In their view, "near Opus" is fair, but "past Opus" is not.
• Very wordy. An independent ranking site found it writes far more than most models to get to an answer. That uses more of a plan's allowance and costs more.
• Always thinking. Its thinking mode is always on. It cannot be switched off.

Moonshot says it removed hacking-related material from parts of its training before the open release. On its own hacking test, the released model scored 81.0%, against 90.5% for a version trained without that removal.

Moonshot says the released model is still behind closed models that have been public for three months or more.

• Open models are catching up, on paper. Strong open models are catching up with the best closed ones, at a lower price. Until others test them, the claims are the maker's own.
• Price per word is not the whole cost. A very wordy model can feel slower and cost more per task, even when its price per word is lower.
• Open cuts both ways. Anyone can run an open model. That is good for choice. It is also a concern for safety, because an open model cannot be called back once it is downloaded.

Midas, the AI assistant in Mythox, does not offer Kimi models right now.

Mythox is waiting for independent tests before deciding whether a new model belongs in Midas. Nothing changes for members today.