diff --git a/.env.example b/.env.example
index 7dc5ed02..dbb60d4b 100644
--- a/.env.example
+++ b/.env.example
@@ -4,9 +4,12 @@ IGNORE_CLIENT_API_KEY=true
 
 OPENAI_BASE_URL=https://api.tokenfactory.nebius.com/v1
 
-BIG_MODEL=zai-org/GLM-4.7-FP8
-MIDDLE_MODEL=zai-org/GLM-4.7-FP8
-SMALL_MODEL=zai-org/GLM-4.7-FP8
+# Nebius rotates model availability — verify IDs against
+#   GET https://api.tokenfactory.nebius.com/v1/models
+# before committing. install.sh does this automatically.
+BIG_MODEL=moonshotai/Kimi-K2.5
+MIDDLE_MODEL=moonshotai/Kimi-K2.5
+SMALL_MODEL=moonshotai/Kimi-K2.5
 VISION_MODEL=Qwen/Qwen2.5-VL-72B-Instruct
 # Optional: explicit context limits (tokens) for safer max_tokens auto-capping
 BIG_MODEL_CONTEXT_LIMIT=204800
@@ -31,4 +34,4 @@ OBSERVABILITY_DB_PATH=/app/data/observability.sqlite3
 OBSERVABILITY_QUEUE_SIZE=1000
 # Keep tool argument storage off unless you explicitly need deeper debugging.
 OBSERVABILITY_STORE_TOOL_ARGS=false
-MODEL_PRICES_JSON='{"zai-org/GLM-4.7-FP8":{"input_per_1m":0.30,"output_per_1m":1.20,"advertised_tok_s":36.8,"currency":"USD"},"Qwen/Qwen2.5-VL-72B-Instruct":{"input_per_1m":0.30,"output_per_1m":1.20,"advertised_tok_s":36.8,"currency":"USD"}}'
+MODEL_PRICES_JSON='{"moonshotai/Kimi-K2.5":{"input_per_1m":0.30,"output_per_1m":1.20,"advertised_tok_s":36.8,"currency":"USD"},"Qwen/Qwen2.5-VL-72B-Instruct":{"input_per_1m":0.30,"output_per_1m":1.20,"advertised_tok_s":36.8,"currency":"USD"}}'