Blog

Notes from the team

Updates, thinking, and technical deep-dives from the Lilac team.

Jul 15, 20262 min read

Lilac partners with Saturn Cloud to expand GPU capacity for Token Factory

Saturn Cloud can scale Token Factory model serving and per-token inference on Lilac capacity without requiring customers to reserve new hardware.

May 7, 20263 min read

We're partnering with MiniMax to bring M2.7 to Lilac

We are partnering with MiniMax to bring commercially licensed MiniMax M2.7 access to Lilac.

May 7, 20262 min read

How to keep frontier open weights viable

Why Lilac supports open-weight licensing, and why commercial rights can help more frontier models stay open.

May 1, 20261 min read

Kimi K2.6 is live on Lilac

Kimi K2.6 is now available on Lilac with OpenAI-compatible chat completions, 262K context, cache-read pricing, and no commitments.

Apr 27, 20262 min read

Cache read pricing is now live on Lilac

Supported Lilac models now show lower cache read rates for repeated context, making long-context and agent workloads cheaper to run.

Apr 8, 20262 min read

Lilac is now self-serve — plus GLM 5.1 and Gemma 4 are live

No more waitlist. Sign up, grab an API key, and start running inference. GLM 5.1 is live at $0.90/M input, and Gemma 4 is live at $0.11/M input.

Apr 8, 20262 min read

GLM 5.1 Inference Benchmark

We benchmarked our GLM 5.1 endpoint against every GLM 5.1 provider listed on OpenRouter. Competitive throughput at the lowest per-token price in the comparison.

Mar 25, 20262 min read

How Idle GPUs Make Cheap Inference Possible

Lilac serves Kimi K2.6 inference on idle enterprise GPUs with OpenAI-compatible, pay-per-token shared endpoints.

Mar 23, 20263 min read

GPU Inference API Pricing Compared

A direct comparison of GPU inference API pricing across major providers. How idle GPU economics enable Lilac to offer lower per-token rates.

Mar 16, 20266 min read

The GPU Scarcity Paradox

The GPU shortage isn't what you think. The industry doesn't have a supply problem — it has a utilization problem masquerading as one.

Mar 1, 20252 min read

Introducing Lilac: Turn Idle GPU Capacity into Revenue

Most Kubernetes clusters run GPUs at 30-50% utilization. We built a single operator to change that.