Moonshot's Kimi K3.1 Pro claims to beat Opus 5.5 and GPT-6 Astra; China's open-model push gains, pending independent tests

Moonshot AI, the Chinese company behind the Kimi models, has launched Kimi K3.1 Pro. The short note we have says it has 2.8 trillion parameters, the adjustable internal values a model learns during training. It also says the model beats Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra on various performance tests. The source does not list which tests, the scores, or whether they were run independently. The note does not say whether the model's weights are publicly available.

Some context helps. In our own rankings, Moonshot's previous model, Kimi K2.6, scored 68 on the Zendoric Quality index, behind the leading Western models and other Chinese rivals such as GLM-5.2 and Qwen3.7-Max. A jump to the very top would be a big leap. That is possible, but it should be shown and not just announced. A model's size alone doesn't decide quality: a larger parameter count does not guarantee a better model.

Our reading: the trend is real. China keeps closing the gap with the frontier, and its open-weight models have been the main engine of that. But "beats X on various tests" is often marketing language: companies choose the benchmarks that flatter them. Before accepting the claim, we will check results on harder, less saturated tests such as SWE-bench Pro and Terminal-Bench, and in independent evaluations. If the results hold up, it is good news overall: more competition at the top pushes prices down and puts powerful tools in more hands. If they don't, it is one more case of the announcement getting ahead of the evidence.