Handle price, responses of models served via Groq

Their tool call response may not strictly follow expected response
format. Let researcher handle incorrect arguments to code tool (i.e
triggers type error)

src/khoj/processor/conversation/openai/utils.py

@@ -934,6 +934,9 @@ async def astream_thought_processor(
             if not chunk_data.get("object") or chunk_data.get("object") != "chat.completion.chunk":
                 logger.warning(f"Skipping invalid chunk with object field: {chunk_data.get('object', 'missing')}")
                 continue
+            # Handle unsupported service tiers like "on_demand" by Groq
+            if chunk.service_tier and chunk.service_tier == "on_demand":
+                chunk_data["service_tier"] = "auto"
 
             tchunk = ChatCompletionWithThoughtsChunk.model_validate(chunk_data)
 

src/khoj/routers/research.py

@@ -433,7 +433,7 @@ async def research(
                         this_iteration.codeContext = code_results
                 async for result in send_status_func(f"**Ran code snippets**: {len(this_iteration.codeContext)}"):
                     yield result
-            except ValueError as e:
+            except (ValueError, TypeError) as e:
                 this_iteration.warning = f"Error running code: {e}"
                 logger.warning(this_iteration.warning, exc_info=True)
 

src/khoj/utils/constants.py

@@ -71,4 +71,7 @@ model_to_cost: Dict[str, Dict[str, float]] = {
     "grok-3-latest": {"input": 3.0, "output": 15.0},
     "grok-3-mini": {"input": 0.30, "output": 0.50},
     "grok-3-mini-latest": {"input": 0.30, "output": 0.50},
+    # Groq pricing
+    "moonshotai/kimi-k2-instruct": {"input": 1.00, "output": 3.00},
+    "openai/gpt-oss-120b": {"input": 0.15, "output": 0.75},
 }