Instructions to use experiential-labs/coding-router with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use experiential-labs/coding-router with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download experiential-labs/coding-router --local-dir coding-router
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Benchmarks: shipped-artifact live per-source results + savings column
Browse files
README.md
CHANGED
|
@@ -35,22 +35,21 @@ At inference: embed the request, compare it against a bank of tasks with measure
|
|
| 35 |
for every model, and pick the cheapest model predicted to solve it. If nothing in the bank
|
| 36 |
is similar to the request, the router does not guess — it escalates to the strongest model.
|
| 37 |
|
| 38 |
-
<!-- Benchmarks
|
| 39 |
-
|
| 40 |
|
| 41 |
## Benchmarks
|
| 42 |
|
| 43 |
-
|
| 44 |
-
|
| 45 |
|
| 46 |
-
|
|
| 47 |
-
|---|---|---|
|
| 48 |
-
|
|
| 49 |
-
|
|
| 50 |
-
|
|
| 51 |
|
| 52 |
-
|
| 53 |
-
cost**, and beats committing to any single model chosen up front on both quality and cost.
|
| 54 |
|
| 55 |
## Scope
|
| 56 |
|
|
|
|
| 35 |
for every model, and pick the cheapest model predicted to solve it. If nothing in the bank
|
| 36 |
is similar to the request, the router does not guess — it escalates to the strongest model.
|
| 37 |
|
| 38 |
+
<!-- Benchmarks section format set by Kion (2026-08-03): the shipped artifact's live-measured
|
| 39 |
+
per-source results, with a savings column. Plain model names, no protocol jargon. -->
|
| 40 |
|
| 41 |
## Benchmarks
|
| 42 |
|
| 43 |
+
Live-measured on held-out tasks the router never saw during fitting, against always using
|
| 44 |
+
opus-5 (high), the strongest single model:
|
| 45 |
|
| 46 |
+
| task type | router quality | opus-5 quality | cheaper | saved |
|
| 47 |
+
|---|---|---|---|---|
|
| 48 |
+
| short interactive requests | 1.000 | 1.000 | **6.0×** | **83%** |
|
| 49 |
+
| competitive programming (LiveCodeBench) | 0.961 | 0.997 | **6.0×** | **83%** |
|
| 50 |
+
| long-horizon software tasks (DeepSWE) | 0.908 | 0.926 | **3.8×** | **74%** |
|
| 51 |
|
| 52 |
+
Quality differences are within statistical noise on every task type; the saving is the win.
|
|
|
|
| 53 |
|
| 54 |
## Scope
|
| 55 |
|