From 551b2a145d85b5b44bf4b9f1b669590d15fa8221 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Sun, 27 Sep 2026 05:44:57 +0000
Subject: [PATCH] Add longcat-2.5-preview-free scores to model-scores.yaml

Model released 2026-09-25 with no published benchmarks (no AA, no LiveBench).
Scores are conservative estimates based on longcat-2.0 baseline with
free-tier reduction applied.

Closes #55
---
 config/model-scores.yaml | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/config/model-scores.yaml b/config/model-scores.yaml
index 032041e..a9a2222 100644
--- a/config/model-scores.yaml
+++ b/config/model-scores.yaml
@@ -77,6 +77,12 @@ model_scores:
     overall: 66.0
     reasoning: 72.0
     vision: 62.0
+  longcat-2.5-preview-free:
+    coding: 66.0             # no benchmarks published (released 2026-09-25); estimate: vendor claims improved coding over longcat-2.0, free tier reduction
+    instruction_following: 86.0  # estimate: longcat-2.0 IF 90, free tier reduction
+    overall: 55.0             # estimate: no LiveBench/AA data; conservative for preview free tier
+    reasoning: 84.0           # estimate: longcat-2.0 reasoning 89, free tier reduction
+    vision: 30.0              # vendor changelog claims image input; no vision benchmark data; conservative
   laguna-s-2.1-free:
     coding: 59.0              # SWE-bench Pro 59.4 (poolside/HF/BenchLM agree); T-Bench 70.2, SWE-Multi 78.5
     instruction_following: 54.0