Kimi K3.1 Pro launched with high token context

On October 23, 2026, Moonshot AI introduced the Kimi K3.1 Pro model, the successor to the Kimi K3 model that was launched in July. This new model features an impressive context window of 1 million tokens and costs approximately one third less via the API than its predecessor. It is currently available in the Kimi apps and via the API, while the open weights are expected on November 4, 2026.

Kimi K3.1 Pro is, according to Moonshot, the first model from the new K3.1 series and offers significantly more computing power for post-processing. With a mixture-of-experts architecture containing 2.8 trillion parameters, 104 billion parameters are active per request. This also makes the model suitable for multimodal applications, including working with images. The test results show that K3.1 Pro achieves a score of 65.7 percent on Terminal-Bench 4.0, placing it near Claude Opus 5.5 and GPT-6 Astra.

The benchmarks highlight that frequent incorrect claims have decreased from 23.4 percent with K3 to 8.3 percent with K3.1 Pro. This indicates improvements in the model's accuracy. An independent evaluation by Artificial Analysis ranks Kimi K3.1 Pro fourth out of 215 models with a score of 55 on their Intelligence Index. For the API, the costs are set at 2 dollars per million input tokens and 10 dollars per million output tokens, making the new K3.1 Pro economically attractive for users.

In addition to K3.1 Pro, Moonshot also has plans for a Highspeed variant expected to be two to three times faster. For privacy, there are various guidelines depending on use: users of the Kimi app must be aware of the storage of personal data in China, while API users have the option for Zero Data Retention under the Singapore data policy. The exact license for this extensive model version has yet to be confirmed, but it is expected to become clear on November 4, 2026.

Read the full article from AI Insider.