Organisation page: Myles-G 12B under official JevBench evaluation (12 Oct 2026)

README.md

@@ -16,8 +16,8 @@ Decision models, built in the UK. Each Myles model answers typed decisions (yes
 | [Myles-22M](https://huggingface.co/launchables/myles-22m) | A browser or a phone | Gated, available |
 | [Myles-150M](https://huggingface.co/launchables/myles-150m) | A laptop or a server CPU | Gated, available |
 | [Myles-400M](https://huggingface.co/launchables/myles-400m) | The cheapest server you have | Gated, available |
-| Myles-G 31B, 12B and Edge | One GPU you own | In preparation |
+| Myles-G 31B, 12B and Edge | One GPU you own | Under evaluation |
 
-Myles-G 31B answers 214 of the 231 public JevBench decisions correctly (92.6%). Unverified: our own measurement, 11 October 2026; official evaluation pending.
+Myles-G 12B is under official evaluation on JevBench, an independent benchmark for decision models (evaluation requested 12 October 2026). Results will be published here once they are official.
 
 [Benchmarks](https://www.launchables.ai/benchmarks) · [Documentation](https://www.launchables.ai/documentation) · [Contact](https://www.launchables.ai/contact)