nebius provider for deepseek

CLAUDE.md

@@ -133,7 +133,7 @@ prefixes determine routing:
 - **Google**: gemini-4-argon, gemini-3.8-flash, gemini-3.7-flash, gemini-3.6-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-2.5-flash, gemini-2.5-flash-lite, gemini-2.5-pro, gemini-3-flash-preview, gemini-3.1-pro-preview; image generation: gemini-2.5-flash-image, gemini-3.1-flash-image, gemini-nano-banana-2.1
 - **xAI**: grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-4.20-reasoning, grok-4.20-non-reasoning; image generation: grok-2-image, grok-imagine-image-2.0
 - **ByteDance** (BytePlus ModelArk, OpenAI-compatible, ap-southeast): seed-1.6, seed-1.8, seed-2.0-lite, deepseek-v4-flash, glm-5.2 (Z.ai's model served via a ModelArk deployment endpoint); image generation: seedream-4.0, seedream-5.0-lite, seedance-4.5, seedance-5.0
-- **OpenRouter** (OpenAI-compatible): deepseek-v4-pro (pinned to Together), hy4-preview, hermes-4-405b, hy3
+- **OpenRouter** (OpenAI-compatible): deepseek-v4-pro (pinned to Nebius FP8), hy4-preview, hermes-4-405b, hy3
 - **abliteration.ai** (OpenAI-compatible): abliterated-model-large-v2
 - **Z.ai** (Model API, OpenAI-compatible): image generation: glm-image (glm-5.2 chat is routed through BytePlus ModelArk, see ByteDance above)
 - **WaveSpeed** (async prediction API): image generation: krea-3 (Krea's model, always `berry` at `reasoning_level: medium`), qwen-image-3.0-pro (Alibaba's model); each is two WaveSpeed endpoints, text-to-image and edit

tee_gateway/model_registry.py

@@ -140,8 +140,8 @@ class ModelConfig:
     # function tools together (preserving reasoning rather than disabling it).
     # OpenAI provider only; ignored elsewhere.
     responses_api_for_tools: bool = False
-    # OpenRouter provider slug to pin the model to (fallbacks off). ``None``
-    # routes to the cheapest provider.
+    # OpenRouter provider slug, or endpoint tag such as ``nebius/fp8``, to pin
+    # the model to (fallbacks off). ``None`` routes to the cheapest provider.
     openrouter_provider: Optional[str] = None
 
     def image_tier(self, resolution: Optional[str]) -> Optional[ImageResolutionTier]:
@@ -1101,14 +1101,14 @@ class SupportedModel(Enum):
     )
 
     # ── OpenRouter (OpenAI-compatible) ──────────────────────────────────
-    # DeepSeek V4 Pro 0813, routed through OpenRouter, pinned to Together
-    # ($1.32 in / $3.96 out per MTok).
+    # DeepSeek V4 Pro 0813, routed through OpenRouter, pinned to Nebius's FP8
+    # endpoint ($1.32 in / $3.96 out per MTok).
     DEEPSEEK_V4_PRO = ModelConfig(
         provider="openrouter",
         api_name="deepseek/deepseek-v4-pro-0813",
         input_price_usd=Decimal("0.00000132"),
         output_price_usd=Decimal("0.00000396"),
-        openrouter_provider="together",
+        openrouter_provider="nebius/fp8",
     )
     # Tencent Hy4 Preview — 770B-parameter (49B active) open-weight MoE,
     # open-sourced 2026-08-28, aimed at coding agents and long-horizon

tee_gateway/test/test_tee_core.py

@@ -1061,12 +1061,12 @@ def tearDownClass(cls) -> None:
             set_provider_config(cls._saved_config)
         get_chat_model_cached.cache_clear()
 
-    def test_deepseek_v4_pro_pinned_to_together(self):
+    def test_deepseek_v4_pro_pinned_to_nebius_fp8(self):
         model = get_chat_model_cached("deepseek-v4-pro", 0.0, 1024)
         self.assertEqual(model.model_name, "deepseek/deepseek-v4-pro-0813")
         self.assertEqual(
             model.extra_body,
-            {"provider": {"order": ["together"], "allow_fallbacks": False}},
+            {"provider": {"order": ["nebius/fp8"], "allow_fallbacks": False}},
         )
 
     def test_unpinned_model_sorts_by_price(self):
@@ -1116,7 +1116,7 @@ def test_openrouter_uses_its_reasoning_object(self):
         self.assertEqual(model.extra_body["reasoning"], {"effort": "none"})
         self.assertEqual(
             model.extra_body["provider"],
-            {"order": ["together"], "allow_fallbacks": False},
+            {"order": ["nebius/fp8"], "allow_fallbacks": False},
         )
 
     def test_openai_takes_it_as_is(self):

tests/test_pricing.py

@@ -597,7 +597,7 @@ def test_deepseek_v4_pro_resolves(self):
         cfg = get_model_config("deepseek-v4-pro")
         self.assertEqual(cfg.provider, "openrouter")
         self.assertEqual(cfg.api_name, "deepseek/deepseek-v4-pro-0813")
-        self.assertEqual(cfg.openrouter_provider, "together")
+        self.assertEqual(cfg.openrouter_provider, "nebius/fp8")
         self.assertEqual(cfg.input_price_usd, Decimal("0.00000132"))
         self.assertEqual(cfg.output_price_usd, Decimal("0.00000396"))