LM Studio Bionic Adds Moonshot AI's 2.8-Trillion-Parameter Kimi K3 Model

LM Studio has added support for Kimi K3, Moonshot AI's 2.8-trillion-parameter mixture-of-experts model, to its agentic workflow platform Bionic. The model ships with a 1-million-token context window, according to LM Studio's announcement on X.

What's new

Bionic launched days ago as LM Studio's dedicated app for agentic tasks—coding, research, and document or file editing—using open models. The platform lets users run models locally or route heavier tasks to cloud-hosted open-source models when local hardware can't keep up on context length, reasoning depth, or tool-calling performance.

Before today, Bionic's cloud lineup included DeepSeek V4, DeepSeek Pro, GLM-5.2, Kimi K2.6, and Kimi-K2.7-Code. Kimi K3 is now the largest and, according to LM Studio, most capable open-source model available on the platform.

All cloud inference in Bionic runs on US-based servers with Zero Data Retention (ZDR) enabled by default. LM Studio says this means prompts, files, and model outputs are not stored once a request finishes processing.

Pricing

Kimi K3 costs $3 per million input tokens, $0.30 per million cached input tokens, and $15 per million output tokens through Bionic. That's a significant jump from the previously available models, which ranged from roughly $0.95 to $1.74 per million input tokens and $3.48 to $4.50 per million output tokens. Cached input pricing is not available for confirmed comparison across the other models in the source data.

Model card specifics beyond parameter count and context window—benchmark scores, training cutoff date, exact architecture details—were not disclosed by Moonshot AI or LM Studio in this announcement.

Availability

LM Studio Bionic is available now for Mac and Windows.

What this means

The price jump is the story here as much as the model itself. At $15 per million output tokens, Kimi K3 costs roughly 3-4x more than the other open models Bionic already offered, which signals Moonshot AI is pricing K3 closer to frontier proprietary models rather than the typically cheaper open-weight tier. For users, that means K3 access through Bionic makes sense for tasks that genuinely need a 1-million-token context window and the reasoning capacity of a 2.8-trillion-parameter MoE—not as a default swap-in for smaller open models.

The move also reinforces a trend among agentic coding and research tools: rather than forcing users to choose between local privacy and cloud capability, platforms like Bionic are blending both, with ZDR policies serving as the trust layer for cloud routing. Whether Kimi K3's premium pricing holds up against GPT and Claude-tier proprietary alternatives on cost-per-task will determine if this becomes Bionic's go-to option for hard problems or a niche, expensive fallback.