Loading post…
Together AIAnnouncement
Together AI announces strategic partnership with Moonshot AI to natively serve Kimi models
Summary
Together AI announces a strategic partnership with Moonshot AI to serve Kimi models as a launch platform, beginning with Kimi K3 and extending to future open-weights releases. Kimi K3 is described as a 2.8-trillion-parameter sparse Mixture-of-Experts model with native vision and a 1M-token context window, using Kimi Delta Attention, Attention Residuals, per-head Muon, and quantile-based expert load balancing. Together says developers can access the models through Serverless, Provisioned Throughput, and Dedicated Inference, with US-hosted infrastructure, zero data retention, autoscaling, and a 99% uptime SLA for Provisioned Throughput. The partnership also supports post-training through full-weight and LoRA reinforcement learning and supervised fine-tuning, with checkpoints deployable directly to inference; Kimi K3 is available at launch, while future Moonshot releases are planned for the same platform.
Context
The partnership positions Together AI as a launch platform for Moonshot AI’s open model releases. It starts with Kimi K3 and is intended to provide day-zero access, production inference, and post-training through one platform and API.
Approach / What changed
Together AI serves Kimi models through Serverless, Provisioned Throughput, and Dedicated Inference products. Developers can post-train models using full-weight or LoRA reinforcement learning and supervised fine-tuning through the Python SDK, then deploy completed checkpoints directly to inference without a separate handoff.
Takeaways
Kimi K3 combines a 2.8-trillion-parameter sparse Mixture-of-Experts design, native vision, and a 1M-token context window with Kimi Delta Attention and Attention Residuals.
Together AI offers Kimi models through Serverless, Provisioned Throughput, and Dedicated Inference, with Provisioned Throughput providing reserved capacity and a 99% uptime SLA.
Post-training supports full-weight and LoRA reinforcement learning plus supervised fine-tuning, and finished checkpoints can move directly into Together AI inference.
FAQ
Which Together AI products can serve Kimi models?
Kimi models are available through Serverless, Provisioned Throughput, and Dedicated Inference. These options cover experimentation, reserved capacity with an SLA, and greater control over production deployment.
Can developers fine-tune Kimi models and deploy the result?
Yes. Developers can use full-weight or LoRA reinforcement learning and advanced supervised fine-tuning through the Python SDK, then deploy a ready checkpoint directly to inference without rebuilding or a separate handoff.
What makes Kimi K3 technically distinctive?
Kimi K3 is a 2.8-trillion-parameter sparse Mixture-of-Experts model with native vision and a 1M-token context window. Its architecture includes Kimi Delta Attention, Attention Residuals, per-head Muon, and quantile-based expert load balancing.
Source and freshness
- Source
- together.ai
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.