August 31: live cluster lane is Vision-Exp. See the recipe post.
| Node | Endpoint | Model | Role |
|---|---|---|---|
| Spark 1 + Spark 2 TP=2 cluster | :8888 | DeepSeek-V4-Flash-Vision-Exp (Mia Anemll recipe, native image, DSpark) | Default local agent lane — 1M context, native vision, KV 1.74M tokens |
| M5 Max oMLX | :8021 | qwen38-27b-ablit-oq4e-mtp | Worker lane — thinking off, MTP, 262K context |
| M5 Max MLX | :8011 | Qwen3-VL-30B (+ model pool) | Vision lane + utility models |
| M5 Max MLX | :8002 / :8003 | Qwen3-Embedding-8B / Qwen3-Reranker-4B | Fleet retrieval |
| M3 Ultra voice | :8768 | will-v8 | Reachy Mini voice lane; Milo-Ark model archive parked on disk |
The big change since the August 25 snapshot: the cluster lane is back on DeepSeek-V4-Flash — now Vision-Exp within 72 hours of the model's release, working through four community recipe generations (SGLang FP8 → NVFP4 DFlash2 → EXL3 900K → EXL3 1M + cache knobs). The full measurement story is in the NVFP4 vs EXL3 faceoff; the current lane's numbers, measured August 29 on the live endpoint:
| Probe (cluster lane, live) | Result |
|---|---|
| Context / KV pool | 1,000,000 tokens / 1,242,424-token pool |
| Cold prefill (32K, nonce-cold) | 886 tok/s |
| Decode structured / prose | 66 / 18.6 tok/s |
| 68K session, turn-2 TTFT (prefix cache) | 1.2 s (~98% hit; was ~75 s) |
| Short request behind a 100K cold prefill | 5.3 s (was ~110 s) |
| Needle retrieval / tools / vision | HIT at 199,970 tokens · 20/20 · pass |
M5 worker/vision/retrieval and M3 voice figures were last benchmarked August 25 (prior snapshot lineage); those endpoints were re-verified alive today but not re-benchmarked. Spark 2 carries no standalone port — it is the TP worker.
This fleet gets the blog posts, but the cloud gets the work. Roughly 95% of our tokens run through four subscriptions:
| Subscription | ~Cost | What it actually does here |
|---|---|---|
| Nous | $200/mo | Family agent fleet — the per-seat economics work well for multiple household agents |
| Anthropic Max | $200/mo | Fable-class models for the heavy agent work; the quota goes fast (this post is being written on it) |
| SuperGrok Heavy | discounted | The daily driver for most ops — best value of the four right now |
| ChatGPT Pro 20× | dropping a tier next cycle | Underused at the current tier; downgrading to match actual consumption |
Why keep the local fleet at all, then? Three real reasons and one honest one. The real ones: privacy lanes (family agents and anything that shouldn't leave the LAN), the robot (voice round-trips to the cloud feel wrong and cost adds up at conversation rates), and retrieval (embeddings and reranking are free and instant at LAN latency). The honest one: running frontier-adjacent open weights on hardware we own is the R&D that keeps us fluent in how these systems actually work — the quant faceoffs, the KV-geometry bugs, the recipe archaeology. That fluency pays for itself when the cloud bill needs negotiating down, which is exactly what's happening with the ChatGPT tier.
Cluster-lane figures measured August 29, 2026 against the live :8888 endpoint (method and artifacts in the faceoff post). M5/M3 endpoints verified alive August 29 via /v1/models and health probes; their throughput figures date to the August 25 session. Cloud percentages are an operational estimate from subscription usage, not token-metered. Journey: June snapshot · GLM-5.3-Flash arrival · NVFP4 vs EXL3 faceoff.