sync: 167 file da Baida98/AI@a58fbf7e (2026-08-22 09:03 UTC)

api/exec.py

@@ -568,15 +568,14 @@ async def llm_fix_code(
     _FIX_CHAIN = []
     groq_key = os.getenv('GROQ_API_KEY', '')
     if groq_key:
-        _FIX_CHAIN.append(('https://api.groq.com/openai/v1', groq_key, 'openai/gpt-oss-120b'))
-        _FIX_CHAIN.append(('https://api.groq.com/openai/v1', groq_key, 'openai/gpt-oss-20b'))
+        _FIX_CHAIN.append(('https://api.groq.com/openai/v1', groq_key, 'qwen/qwen3.6-27b'))
 
     or_key = os.getenv('OPENROUTER_API_KEY', '')
     if or_key:
         for m in [
-            'openai/gpt-oss-20b:free',
-            'meta-llama/llama-3.1-8b-instruct:free',
-            'qwen/qwen-2.5-coder-7b-instruct:free',
+            'openrouter/free',
+            'qwen/qwen3-coder:free',
+            'meta-llama/llama-3.3-70b-instruct:free',
         ]:
             _FIX_CHAIN.append(('https://openrouter.ai/api/v1', or_key, m))
 

api/providers.py

@@ -930,14 +930,14 @@ async def update_provider_models(role: AuthRole = Depends(require_role(AuthRole.
     # filtrare per `name` evita di applicare un formato incompatibile alla riga sbagliata.
     # GPT-OSS 120B non compare come vecchio valore perché è già un modello supportato.
     MODEL_FIXES = [
-        ("groq",       "llama-3.1-70b-versatile", "openai/gpt-oss-120b"),
+        ("groq",       "llama-3.1-70b-versatile", "qwen/qwen3.6-27b"),
         ("cerebras",   "llama3.1-70b",            "gpt-oss-120b"),
         ("nvidia",     "llama-3.1-405b-instruct", "meta/llama-3.3-70b-instruct"),
-        ("openrouter", "llama-3.1-405b",          "openai/gpt-oss-20b:free"),
+        ("openrouter", "llama-3.1-405b",          "openrouter/free"),
         ("sambanova",  "llama3-70b",              "DeepSeek-V3.2"),
         ("gemini",     "gemini-1.5-flash",        "gemini-3.5-flash-lite"),
         ("gemini",     "gemini-1.5-pro",          "gemini-3.6-flash"),
-        ("openrouter", "claude-3.5-sonnet",       "openai/gpt-oss-20b:free"),
+        ("openrouter", "claude-3.5-sonnet",       "openrouter/free"),
     ]
 
     import asyncio as _aio

models/ai_client.py

@@ -61,11 +61,11 @@ class ProviderConfig:
 # (nessun proxy CF Worker qui: questo client gira lato backend Python, non browser).
 _PROVIDER_DEFS = [
     # tier 0 — free tier veloce e affidabile
-    {"name": "groq",       "env_key": "GROQ_API_KEY",       "base_url": "https://api.groq.com/openai/v1",                    "model_env": "GROQ_MODEL",      "default_model": "openai/gpt-oss-120b",     "tier": 0, "purpose": "reasoning"},
+    {"name": "groq",       "env_key": "GROQ_API_KEY",       "base_url": "https://api.groq.com/openai/v1",                    "model_env": "GROQ_MODEL",      "default_model": "qwen/qwen3.6-27b",     "tier": 0, "purpose": "reasoning"},
     {"name": "cerebras",   "env_key": "CEREBRAS_API_KEY",   "base_url": "https://api.cerebras.ai/v1",                        "model_env": "CEREBRAS_MODEL",  "default_model": "gpt-oss-120b",                "tier": 0, "purpose": "reasoning"},
     {"name": "sambanova",  "env_key": "SAMBANOVA_API_KEY",  "base_url": "https://api.sambanova.ai/v1",                       "model_env": "SAMBANOVA_MODEL", "default_model": "DeepSeek-V3.1",               "tier": 0, "purpose": "reasoning"},
     # tier 1 — free tier con rate limit più stretti
-    {"name": "openrouter", "env_key": "OPENROUTER_API_KEY", "base_url": "https://openrouter.ai/api/v1",                      "model_env": "OPENROUTER_MODEL","default_model": "openai/gpt-oss-20b:free", "tier": 1, "purpose": "coding"},
+    {"name": "openrouter", "env_key": "OPENROUTER_API_KEY", "base_url": "https://openrouter.ai/api/v1",                      "model_env": "OPENROUTER_MODEL","default_model": "openrouter/free", "tier": 1, "purpose": "coding"},
     {"name": "hf_router",  "env_key": "HF_TOKEN",           "base_url": "https://router.huggingface.co/v1",                  "model_env": "HF_MODEL",        "default_model": "Qwen/Qwen2.5-Coder-32B-Instruct", "tier": 1, "purpose": "coding"},
     {"name": "gemini",     "env_key": "GEMINI_API_KEY",     "base_url": "https://generativelanguage.googleapis.com/v1beta/openai/", "model_env": "GEMINI_MODEL", "default_model": "gemini-3.6-flash",        "tier": 1, "purpose": "memory"},
     # tier 2 — fallback opzionale (spesso a pagamento o quota limitata)

models/role_router.py

@@ -3,27 +3,27 @@ role_router.py — Multi-model role routing (S362, aggiornato 2026-06-14 benchma
 
 BENCHMARK RESULTS 2026-06-14 FINALE (14 modelli × 3 test, max_tokens corretti):
   100% qualità (ordinati per TTFT):
-  #1  Groq / openai/gpt-oss-20b           — 170ms  100% ← FASTEST
+  #1  Groq / qwen/qwen3.6-27b           — 170ms  100% ← FASTEST
   #2  Cerebras / gpt-oss-120b               — 207ms  100% ← REASONING (max_tokens≥500)
-  #3  Groq / openai/gpt-oss-120b        — 235ms  100%
+  #3  Groq / qwen/qwen3.6-27b        — 235ms  100%
   #4  Cerebras / zai-glm-4.7               — 254ms  100%
   #5  Groq / compound-mini                  — 341ms  100%
   #6  SambaNova / DeepSeek-V3.1            — 482ms  100%
   #7  SambaNova / gemma-4-31B              — 2132ms 100%
-  #8  OpenRouter / gpt-oss-120b:free        — 2160ms 100%
+  #8  OpenRouter / openrouter/free        — 2160ms 100%
 
 Role assignments 2026-06-14 FINALE:
-  FAST       → Groq openai/gpt-oss-20b      (170ms, 100%) ← #1 assoluto
+  FAST       → Groq qwen/qwen3.6-27b      (170ms, 100%) ← #1 assoluto
   ARCHITECT  → Groq llama-4-scout-17b 10M ctx  (244ms, 67%  — best per contesto lungo)
-  CODER      → Groq openai/gpt-oss-120b    (235ms, 100%) ← #3 qualità
-  TESTER     → Groq openai/gpt-oss-20b
-  CONTEXT    → Groq openai/gpt-oss-20b
+  CODER      → Groq qwen/qwen3.6-27b    (235ms, 100%) ← #3 qualità
+  TESTER     → Groq qwen/qwen3.6-27b
+  CONTEXT    → Groq qwen/qwen3.6-27b
   RESEARCHER → Gemini 2.5-flash (599ms, 67% — math prompt-sensitive)
   REASONER   → Cerebras gpt-oss-120b (207ms, 100%, reasoning model → max_tokens≥500)
   SAMBANOVA  → SambaNova DeepSeek-V3.1 (482ms, 100%)
-  DEFAULT    → AIClient() primary (openai/gpt-oss-120b o primo disponibile)
+  DEFAULT    → AIClient() primary (qwen/qwen3.6-27b o primo disponibile)
 
-OpenRouter tenuto come fallback secondario (gpt-oss-120b:free = 1645ms ma 100% qualità).
+OpenRouter tenuto come fallback secondario (openrouter/free = 1645ms ma 100% qualità).
 """
 from __future__ import annotations
 
@@ -36,11 +36,11 @@ _logger = logging.getLogger("models.role_router")
 
 
 class Role(str, Enum):
-    FAST       = "fast"        # greetings, math semplice, identity — openai/gpt-oss-20b
+    FAST       = "fast"        # greetings, math semplice, identity — qwen/qwen3.6-27b
     ARCHITECT  = "architect"   # planning, ragionamento complesso — GPT-OSS 120B
-    CODER      = "coder"       # coding, debug — openai/gpt-oss-120b
-    TESTER     = "tester"      # test gen, debug hints — openai/gpt-oss-120b
-    CONTEXT    = "context"     # summarization, context compression — openai/gpt-oss-120b
+    CODER      = "coder"       # coding, debug — qwen/qwen3.6-27b
+    TESTER     = "tester"      # test gen, debug hints — qwen/qwen3.6-27b
+    CONTEXT    = "context"     # summarization, context compression — qwen/qwen3.6-27b
     DEFAULT    = "default"     # AIClient() primary
     RESEARCHER = "researcher"  # web research + document synthesis — GPT-OSS 120B
     REASONER   = "reasoner"    # throughput massimo — Cerebras GPT-OSS 120B
@@ -122,7 +122,7 @@ class RoleRouter:
             name="groq-fast",
             api_key=groq_key,
             base_url="https://api.groq.com/openai/v1",
-            default_model=os.getenv("GROQ_FAST_MODEL", "openai/gpt-oss-20b"),
+            default_model=os.getenv("GROQ_FAST_MODEL", "qwen/qwen3.6-27b"),
         )
         rest = [p for p in client.providers if p.name not in ("groq", "groq-fast", "groq-tester")]
         client.providers = [fast, *rest]
@@ -163,7 +163,7 @@ class RoleRouter:
                 name="groq-architect",
                 api_key=groq_key,
                 base_url="https://api.groq.com/openai/v1",
-                default_model=os.getenv("ARCHITECT_MODEL", "openai/gpt-oss-120b"),
+                default_model=os.getenv("ARCHITECT_MODEL", "qwen/qwen3.6-27b"),
             )
             rest = [p for p in client.providers if p.name not in ("groq", "groq-architect")]
             client.providers = [architect, *rest]
@@ -179,7 +179,7 @@ class RoleRouter:
                 profiles = [ProviderConfig(
                     name="openrouter", api_key=os.getenv("OPENROUTER_API_KEY", ""),
                     base_url="https://openrouter.ai/api/v1",
-                    default_model=os.getenv("OPENROUTER_MODEL", "openai/gpt-oss-20b:free"),
+                    default_model=os.getenv("OPENROUTER_MODEL", "openrouter/free"),
                     profile="legacy",
                 )]
             if profiles:
@@ -201,7 +201,7 @@ class RoleRouter:
             return profiled
         groq_key = os.getenv("GROQ_API_KEY")
         nvidia_key = os.getenv("NVIDIA_API_KEY")
-        model = os.getenv("CODER_MODEL", "openai/gpt-oss-120b")
+        model = os.getenv("CODER_MODEL", "qwen/qwen3.6-27b")
         if groq_key:
             client = AIClient()
             coder = ProviderConfig(
@@ -241,7 +241,7 @@ class RoleRouter:
                 profiles = [ProviderConfig(
                     name="openrouter", api_key=os.getenv("OPENROUTER_API_KEY", ""),
                     base_url="https://openrouter.ai/api/v1",
-                    default_model=os.getenv("OPENROUTER_MODEL", "openai/gpt-oss-20b:free"),
+                    default_model=os.getenv("OPENROUTER_MODEL", "openrouter/free"),
                     profile="legacy",
                 )]
             if profiles:
@@ -273,7 +273,7 @@ class RoleRouter:
                 base_url="https://api.groq.com/openai/v1",
                 default_model=os.getenv(
                     "GROQ_RESEARCH_MODEL",
-                    os.getenv("GROQ_MODEL", "openai/gpt-oss-120b"),
+                    os.getenv("GROQ_MODEL", "qwen/qwen3.6-27b"),
                 ),
             )
             rest = [
@@ -307,7 +307,7 @@ class RoleRouter:
                 base_url="https://api.groq.com/openai/v1",
                 default_model=os.getenv(
                     "GROQ_REASONER_MODEL",
-                    os.getenv("GROQ_MODEL", "openai/gpt-oss-120b"),
+                    os.getenv("GROQ_MODEL", "qwen/qwen3.6-27b"),
                 ),
             )
             rest = [
@@ -406,7 +406,7 @@ class RoleRouter:
             name="groq-tester",
             api_key=groq_key,
             base_url="https://api.groq.com/openai/v1",
-            default_model=os.getenv("GROQ_FAST_MODEL", "openai/gpt-oss-20b"),
+            default_model=os.getenv("GROQ_FAST_MODEL", "qwen/qwen3.6-27b"),
             purpose="coding",
         )
         rest = [p for p in client.providers if p.name not in ("groq", "groq-tester")]

tests/test_provider_model_migrations.py

@@ -47,7 +47,7 @@ class RoleRouterModelMigrationTests(unittest.TestCase):
         client = RoleRouter.get_client(Role.ARCHITECT)
 
         self.assertEqual(client.provider_name, "groq-architect")
-        self.assertEqual(client.default_model, "openai/gpt-oss-120b")
+        self.assertEqual(client.default_model, "qwen/qwen3.6-27b")
 
     @patch.object(AIClient, "_load_providers", return_value=[])
     @patch.dict(
@@ -59,8 +59,8 @@ class RoleRouterModelMigrationTests(unittest.TestCase):
         architect = RoleRouter.get_client(Role.ARCHITECT)
         coder = RoleRouter.get_client(Role.CODER)
 
-        self.assertEqual(architect.default_model, "openai/gpt-oss-20b:free")
-        self.assertEqual(coder.default_model, "openai/gpt-oss-20b:free")
+        self.assertEqual(architect.default_model, "openrouter/free")
+        self.assertEqual(coder.default_model, "openrouter/free")
 
     @patch.object(AIClient, "_load_providers", return_value=[])
     @patch.dict(
@@ -90,14 +90,14 @@ class ProviderTableMigrationTests(unittest.IsolatedAsyncioTestCase):
         self.assertEqual(len(fake_supabase.table_ref.calls), 8)
 
         expected = {
-            ("groq", "llama-3.1-70b-versatile", "openai/gpt-oss-120b"),
+            ("groq", "llama-3.1-70b-versatile", "qwen/qwen3.6-27b"),
             ("cerebras", "llama3.1-70b", "gpt-oss-120b"),
             ("nvidia", "llama-3.1-405b-instruct", "meta/llama-3.3-70b-instruct"),
-            ("openrouter", "llama-3.1-405b", "openai/gpt-oss-20b:free"),
+            ("openrouter", "llama-3.1-405b", "openrouter/free"),
             ("sambanova", "llama3-70b", "DeepSeek-V3.2"),
             ("gemini", "gemini-1.5-flash", "gemini-3.5-flash-lite"),
             ("gemini", "gemini-1.5-pro", "gemini-3.6-flash"),
-            ("openrouter", "claude-3.5-sonnet", "openai/gpt-oss-20b:free"),
+            ("openrouter", "claude-3.5-sonnet", "openrouter/free"),
         }
         actual = {
             (