Use gpt-5.4-mini as fallback profile model

Author: WW-shan

docs/superpowers/plans/2026-05-13-llm-research-profile.md

@@ -90,7 +90,7 @@ def test_balanced_profile_exports_benchmark_provider_order():
     assert values["OPENAI_BACKUP_MODEL"] == "longcat-flash-chat"
     assert values["OPENAI_BACKUP_BASE_URL"] == "https://elysiver.h-e.top/v1"
     assert values["OPENAI_BACKUP_API_MODE"] == "chat"
-    assert values["OPENAI_FALLBACK_MODEL"] == "gpt-5.4"
+    assert values["OPENAI_FALLBACK_MODEL"] == "gpt-5.4-mini"
     assert values["OPENAI_FALLBACK_BASE_URL"] == "https://api.wwcloud.app"
     assert values["OPENAI_FALLBACK_API_MODE"] == "responses"
 
@@ -260,7 +260,7 @@ llm_profile_set_role OPENAI_SECONDARY_API_KEY OPENAI_SECONDARY_MODEL OPENAI_SECO
 llm_profile_set_role OPENAI_BACKUP_API_KEY OPENAI_BACKUP_MODEL OPENAI_BACKUP_BASE_URL OPENAI_BACKUP_API_MODE \
   "longcat-flash-chat" "https://elysiver.h-e.top/v1" "chat"
 llm_profile_set_role OPENAI_FALLBACK_API_KEY OPENAI_FALLBACK_MODEL OPENAI_FALLBACK_BASE_URL OPENAI_FALLBACK_API_MODE \
-  "gpt-5.4" "https://api.wwcloud.app" "responses"
+  "gpt-5.4-mini" "https://api.wwcloud.app" "responses"
 
 llm_profile_print_summary
 ```

docs/superpowers/specs/2026-05-13-llm-research-profile-design.md

@@ -36,7 +36,7 @@ Provider order:
 | primary | `windhub/deepseek-v3-2-251201` | `https://windhub.cc/v1` | `messages` | Best balance of strict recall and latency among stable candidates |
 | secondary | `secondary/gemini-3.1-pro-preview` | `https://api.xn--chy-js0fk50c.top/v1` | `chat` | Formal CLI smoke passed; lower semantic strength but fast |
 | backup | `elysiver/longcat-flash-chat` | `https://elysiver.h-e.top/v1` | `chat` | Best stable elysiver result, good recall and moderate latency |
-| fallback | `gpt-5.4` on the original responses endpoint | `https://api.wwcloud.app` | `responses` | More expensive high-capability last resort |
+| fallback | `gpt-5.4-mini` on the original responses endpoint | `https://api.wwcloud.app` | `responses` | Lower-cost responses-compatible last resort |
 
 Optional profile: `semantic`
 

scripts/load_llm_research_profile.sh

@@ -64,6 +64,6 @@ llm_profile_set_role OPENAI_SECONDARY_API_KEY OPENAI_SECONDARY_MODEL OPENAI_SECO
 llm_profile_set_role OPENAI_BACKUP_API_KEY OPENAI_BACKUP_MODEL OPENAI_BACKUP_BASE_URL OPENAI_BACKUP_API_MODE \
   "longcat-flash-chat" "https://elysiver.h-e.top/v1" "chat"
 llm_profile_set_role OPENAI_FALLBACK_API_KEY OPENAI_FALLBACK_MODEL OPENAI_FALLBACK_BASE_URL OPENAI_FALLBACK_API_MODE \
-  "gpt-5.4" "https://api.wwcloud.app" "responses"
+  "gpt-5.4-mini" "https://api.wwcloud.app" "responses"
 
 llm_profile_print_summary

tests/test_llm_research_profile.py

@@ -55,7 +55,7 @@ def test_balanced_profile_exports_benchmark_provider_order():
     assert values["OPENAI_BACKUP_MODEL"] == "longcat-flash-chat"
     assert values["OPENAI_BACKUP_BASE_URL"] == "https://elysiver.h-e.top/v1"
     assert values["OPENAI_BACKUP_API_MODE"] == "chat"
-    assert values["OPENAI_FALLBACK_MODEL"] == "gpt-5.4"
+    assert values["OPENAI_FALLBACK_MODEL"] == "gpt-5.4-mini"
     assert values["OPENAI_FALLBACK_BASE_URL"] == "https://api.wwcloud.app"
     assert values["OPENAI_FALLBACK_API_MODE"] == "responses"