- Who it helps
- Developers with an M5 Max Mac evaluating local model serving.
- Availability
- Public Apache-2.0 code. The README requires a 40-core-GPU M5 Max, 128 GB unified memory and macOS 27 or later; model terms are separate.
A narrow hardware target
The documented setup serves Qwen3.8 Flash Next with a 65,536-token context and a generated access key. The API binds to the local machine by default.
Compare under matched conditions
The launch makes a speed claim, but no matched benchmark was reproduced. The stated machine requirements do not show that smaller Macs can run the model or that concurrency has negligible memory cost.
What you can do
- Check the exact Mac configuration and model licence before planning a local evaluation. Open reference ↗
Where the evidence stops
No build, download or inference executed. The model-licence file could not be read, so deployment rights and full resource costs remain unverified.
Sources read. Public source text read November 10, 2026 UTC. No third-party project, model inference or complete workflow executed.
Sources & further reading
GitHub · implementation
Tokspur local inference repository ↗
README specifies M5 Max, 128 GB unified memory and macOS 27 or later. Model licence file retrieval failed.
san-digital · page checked 2026-11-10
12 stars · observed 2026-11-10 12:24 UTC
More sources reviewed (1)
Hacker News · discussion
New Show HN submissions ↗
First fourteen titles inspected; ZTHR and BlablaMeter followed to primary descriptions.
Hacker News · page checked 2026-11-10
Coverage note: Model-licence file and independent matched benchmarks remain unverified.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
Tokspur serves a local language model through its own Apple Metal inference worker and an HTTP interface.