The Story
Splash is our open-source inference engine for Apple silicon, built around the model rather than around a model zoo. On a 48 GB M5 Pro it generates 210 tokens/s on Qwen3.6-35B-A3B, reopens a cached 32K context in 123 ms, and serves four concurrent requests at 357 tokens/s combined.
Most inference engines are built to run any model for maximum flexibility. Our approach turns this assumption upside down: the engine is built around the model for maximum efficiency . Its kernels, draft model, and memory plan are specialized for the model it serves. This design runs our industry-leading inference platform on data-center GPUs, where we are the fastest provider on Artificial Analysis for five frontier open models.
Today we bring the same technology to Apple silicon with Splash , our open-source inference engine for the Mac. It delivers 2× the decode speed of the next-fastest engine we measured on Qwen3.8-27B and stays ahead at every context length we tested, to 32K tokens. At four parallel subagents, the speedup increases to almost 4× .
Overview
Splash is our open-source inference engine for Apple silicon, built around the model rather than around a model zoo. On a 48 GB M5 Pro it generates 210...
Splash is our open-source inference engine for Apple silicon, built around the model rather than around a model zoo. On a 48 GB M5 Pro it generates 210 tokens/s on Qwen3.6-35B-A3B, reopens a cached 32K context in 123 ms, and serves four concurrent requests at 357 tokens/s combined.
Most inference engines are built to run any model for maximum flexibility. Our approach turns this assumption upside down: the engine is built around the model for maximum efficiency . Its kernels, draft model, and memory plan are specialized for the model it serves. This design runs our industry-leading inference platform on data-center GPUs, where we are the fastest provider on Artificial Analysis for five frontier open models.
Today we bring the same technology to Apple silicon with Splash , our open-source inference engine for the Mac. It delivers 2× the decode speed of the next-fastest engine we measured on Qwen3.8-27B and stays ahead at every context length we tested, to 32K tokens. At four parallel subagents, the speedup increases to almost 4× .
FAQ
What is Splash?
Splash is our open-source inference engine for Apple silicon, built around the model rather than around a model zoo. On a 48 GB M5 Pro it generates 210...
Where can I access Splash?
Use the official product link below. It is the destination recorded for this listing.
Where should I check current pricing and availability?
Check the current product page before making a decision. The directory's pricing category is a listing field, not a live price quote or availability guarantee.
What stack is Splash built with?
The CompassPad entry does not list a specific stack for Splash.
Is Splash free, or does it cost money?
The CompassPad listing records Splash as open source. Check the repository for its current license and usage terms.
What we checked
- DISCOVEREDFound in the wild
This listing was discovered from a public launch post and the description above is drawn from the product's own website.
- LIVEProduct site
The product site resolved when this entry was added to the catalog.
Discussion
No comments yet. Be the first.
Join the conversation. Sign up to comment.
Sign up free