fix(volcengine): support tiered Doubao pricing

litellm/cost_calculator.py

@@ -642,6 +642,12 @@ def cost_per_token(
         )
 
         return dashscope_cost_per_token(model=model, usage=usage_block)
+    elif custom_llm_provider == "volcengine":
+        from litellm.llms.volcengine.cost_calculator import (
+            cost_per_token as volcengine_cost_per_token,
+        )
+
+        return volcengine_cost_per_token(model=model, usage=usage_block)
     elif custom_llm_provider == "azure_ai":
         return azure_ai_cost_per_token(
             model=model,

litellm/llms/volcengine/cost_calculator.py

@@ -0,0 +1,86 @@
+"""
+Cost calculator for Volcengine chat models.
+
+Volcengine selects one pricing tier from the request's total input length and
+charges every input/output token at that tier. This is not graduated pricing.
+"""
+
+from litellm.litellm_core_utils.llm_cost_calc.tiered_pricing import (
+    select_tier_for_input,
+    tier_rate,
+)
+from litellm.types.utils import ModelInfo, Usage
+from litellm.utils import get_model_info
+
+
+def _cached_prompt_tokens(usage: Usage) -> int:
+    prompt_details = usage.prompt_tokens_details
+    if prompt_details is None:
+        return 0
+    return int(getattr(prompt_details, "cached_tokens", 0) or 0)
+
+
+def _select_pricing_tier(
+    tiered_pricing: list[dict] | None,  # mutable-ok: shared tier helper accepts JSON-backed lists of dictionaries
+    prompt_tokens: int,
+) -> dict | None:  # mutable-ok: returns the JSON-backed tier selected by the shared helper
+    if not tiered_pricing:
+        return None
+
+    # Output-only synthetic usage blocks have no input length. Use the first
+    # tier instead of returning zero cost for their completion tokens.
+    return select_tier_for_input(
+        tiered_pricing=tiered_pricing,
+        input_tokens=max(prompt_tokens, 1),
+    )
+
+
+def _output_rate(
+    tier: dict,  # mutable-ok: tier_rate accepts a JSON-backed dictionary
+    completion_tokens: int,
+) -> float:
+    """Read the output rate, including Seed 1.8's short-output discount."""
+    if completion_tokens > 200 and tier.get("output_cost_per_token_above_200_tokens") is not None:
+        return tier_rate(tier, "output_cost_per_token_above_200_tokens")
+    return tier_rate(tier, "output_cost_per_token")
+
+
+def cost_per_token(model: str, usage: Usage) -> tuple[float, float]:
+    """
+    Return ``(prompt_cost_usd, completion_cost_usd)`` for a Volcengine request.
+
+    Tier selection is based on total prompt tokens. Cached prompt tokens still
+    count toward the tier boundary, but use the cache-read rate when one is
+    declared by the model.
+    """
+    model_info: ModelInfo = get_model_info(
+        model=model,
+        custom_llm_provider="volcengine",
+    )
+
+    prompt_tokens = int(usage.prompt_tokens or 0)
+    completion_tokens = int(usage.completion_tokens or 0)
+    cached_tokens = min(_cached_prompt_tokens(usage), prompt_tokens)
+    uncached_tokens = prompt_tokens - cached_tokens
+
+    raw_tiered_pricing = model_info.get("tiered_pricing")
+    tiered_pricing = raw_tiered_pricing if isinstance(raw_tiered_pricing, list) else None
+    tier = _select_pricing_tier(
+        tiered_pricing=tiered_pricing,
+        prompt_tokens=prompt_tokens,
+    )
+
+    if tier is not None:
+        input_rate = tier_rate(tier, "input_cost_per_token")
+        output_rate = _output_rate(tier=tier, completion_tokens=completion_tokens)
+        cache_rate_value = model_info.get("cache_read_input_token_cost")
+        cache_rate = float(cache_rate_value) if cache_rate_value is not None else input_rate
+    else:
+        input_rate = float(model_info.get("input_cost_per_token") or 0.0)
+        output_rate = float(model_info.get("output_cost_per_token") or 0.0)
+        cache_rate_value = model_info.get("cache_read_input_token_cost")
+        cache_rate = float(cache_rate_value) if cache_rate_value is not None else input_rate
+
+    prompt_cost = (uncached_tokens * input_rate) + (cached_tokens * cache_rate)
+    completion_cost = completion_tokens * output_rate
+    return prompt_cost, completion_cost

litellm/model_prices_and_context_window_backup.json

@@ -45281,6 +45281,47 @@
         "supports_vision": true,
         "source": "https://aws.amazon.com/bedrock/pricing/"
     },
+    "volcengine/doubao-seed-1-8-251228": {
+        "cache_read_input_token_cost": 2.2e-08,
+        "litellm_provider": "volcengine",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
+        "mode": "chat",
+        "source": "https://www.volcengine.com/docs/82379/1330310",
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_tool_choice": false,
+        "supports_vision": true,
+        "tiered_pricing": [
+            {
+                "input_cost_per_token": 1.1e-07,
+                "output_cost_per_token": 2.8e-07,
+                "output_cost_per_token_above_200_tokens": 1.1e-06,
+                "range": [
+                    0,
+                    32000.0
+                ]
+            },
+            {
+                "input_cost_per_token": 1.7e-07,
+                "output_cost_per_token": 2.2e-06,
+                "range": [
+                    32000.0,
+                    128000.0
+                ]
+            },
+            {
+                "input_cost_per_token": 3.3e-07,
+                "output_cost_per_token": 3.3e-06,
+                "range": [
+                    128000.0,
+                    256000.0
+                ]
+            }
+        ]
+    },
     "volcengine/doubao-seed-2-0-pro-260215": {
         "litellm_provider": "volcengine",
         "max_input_tokens": 256000,

model_prices_and_context_window.json

@@ -45402,6 +45402,47 @@
         "supports_vision": true,
         "source": "https://aws.amazon.com/bedrock/pricing/"
     },
+    "volcengine/doubao-seed-1-8-251228": {
+        "cache_read_input_token_cost": 2.2e-08,
+        "litellm_provider": "volcengine",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 64000,
+        "max_tokens": 64000,
+        "mode": "chat",
+        "source": "https://www.volcengine.com/docs/82379/1330310",
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_tool_choice": false,
+        "supports_vision": true,
+        "tiered_pricing": [
+            {
+                "input_cost_per_token": 1.1e-07,
+                "output_cost_per_token": 2.8e-07,
+                "output_cost_per_token_above_200_tokens": 1.1e-06,
+                "range": [
+                    0,
+                    32000.0
+                ]
+            },
+            {
+                "input_cost_per_token": 1.7e-07,
+                "output_cost_per_token": 2.2e-06,
+                "range": [
+                    32000.0,
+                    128000.0
+                ]
+            },
+            {
+                "input_cost_per_token": 3.3e-07,
+                "output_cost_per_token": 3.3e-06,
+                "range": [
+                    128000.0,
+                    256000.0
+                ]
+            }
+        ]
+    },
     "volcengine/doubao-seed-2-0-pro-260215": {
         "litellm_provider": "volcengine",
         "max_input_tokens": 256000,