Kimi K3.1 Pro (Inference.net)

Inference.net announced on November 4, 2026 that Kimi K3.1 Pro was available on its platform from day one, enabled by Moonshot publishing open weights. The model card describes a post-trained Kimi K3 variant with 2.8 trillion parameters, roughly one million tokens of context, and low, high and max reasoning levels with max as default. On Inference.net the model is exposed through an OpenAI-compatible API, supports image input and tool calling, and needs no extra setup for prompt caching, with repeated prefixes billed at a lower cache-read rate. Its listing showed $2 per million input tokens and $10 per million output tokens against $3 and $15 for Kimi K3.

Moonshot AI introduced Kimi K3.1 Pro on October 24, 2026 as the first model in its new K3.1 series, describing it as its most capable 3T-class open-weight model to date with a 1-million-token context window. A faster companion variant, Kimi K3.1 Pro Highspeed, delivers two to three times the speed with identical capabilities. Both models are reachable through the Kimi API as kimi-k3.1-pro and kimi-k3.1-pro-highspeed, with Zero Data Retention available for API customers. Reported results include 65.7 percent on Terminal-Bench 4.0, ahead of GPT-6 Astra at 58.2 percent, and 61.2 percent on Kimi Code Bench 3.0 Long-horizon versus 55.95 percent for Kimi K3.

Kimi K3.1 Pro launched on October 23, 2026 as a post-trained 2.8-trillion-parameter sparse mixture-of-experts model with 104 billion active parameters and a 1-million-token context window, enhanced by large-scale joint reinforcement learning. Open weights are published on Hugging Face as moonshotai/Kimi-K3.1-Pro under the Kimi K3.1 Pro License. Reported benchmark results include 65.7 percent on Terminal-Bench 4.0, 92.7 percent on BenchCAD with tools, and a reduction in claim-level hallucinations from 23.4 percent to 8.3 percent. On the Artificial Analysis Intelligence Index it debuted in 4th place with a score of 55, and official API pricing fell roughly 33 percent versus Kimi K3.

Simon Willison tested Kimi K3.1 Pro through Moonshot’s OpenAI-compatible API, confirming that developers could invoke it with an existing client by changing the endpoint and model name. He observed three reasoning settings—low, medium, and max—and compared outputs from the model with those from Kimi K3. Launch pricing was $2 per million input tokens and $10 per million output tokens, while the Highspeed variant cost twice as much except for cache-hit input. Artificial Analysis reportedly scored K3.1 Pro at 55, ahead of several frontier models at 53 but behind Claude Opus 5.5 at 58.

Moonshot’s official quickstart documents Kimi K3.1 Pro as a post-trained Kimi K3 variant with a 1-million-token context window and improved coding, 3D creation, financial research, and agentic capabilities. It is designed for both quick repository fixes and full-project builds, including work with Unity, Unreal Engine, Godot, and Three.js. The documentation reports a 65.7% Terminal-Bench 4.0 score and 61.2% on Kimi Code Bench 3.0 Long-horizon, versus 55.95% for K3. It also identifies Kimi K3.1 Pro Highspeed as a distinct release offering two to three times the standard model’s speed for more responsive interactions.

Moonshot AI introduced Kimi K3.1 Pro as the first model in its K3.1 series and its most capable 3T-class open-weight model to date, with a 1-million-token context window. It is a post-trained version of Kimi K3 that uses substantially more post-training compute, including joint reinforcement learning across hundreds of downstream domains. Moonshot reports that K3.1 Pro scored 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%. It also scored 61.2% on the internal Kimi Code Bench 3.0 Long-horizon, compared with 55.95% for K3, while a separate Highspeed variant was released at two to three times the speed.