@qa-gentic/stlc-agents 1.0.25 → 1.0.26

package/src/stlc_agents/shared/pricing.py

@@ -2,13 +2,28 @@
 pricing.py  —  Model pricing registry for stlc-agents cost tracking.
 
 Prices: USD per million tokens (MTok).
-Source: Anthropic official docs, April 2026.
 
-Models this repo actually calls:
-  - claude-sonnet-4-20250514  (LocatorHealer AI Vision, default)
-  - gpt-4o                    (LocatorHealer AI Vision, copilot provider)
-  + whatever coding agent the user runs (Claude / Copilot / Cursor / Windsurf)
-    — the user declares this via STLC_CODING_AGENT_MODEL env var.
+Sources (verified April 2026):
+  Anthropic  — https://docs.anthropic.com/en/docs/about-claude/models/overview
+               cache_write = 5-min TTL rate; cache_read = 0.1× input rate
+  OpenAI     — https://openai.com/api/pricing  (standard tier)
+               columns: [model, input, cached_input, output]
+               no cache_write fee (caching is automatic)
+  xAI        — https://docs.x.ai/docs/models
+  DeepSeek   — https://api-docs.deepseek.com/quick_start/pricing
+               cache_read = cache-hit input rate (10× cheaper than cache miss)
+
+Column layout for ModelPricing:
+  input_per_mtok      — uncached input tokens (full rate)
+  output_per_mtok     — output / completion tokens
+  cache_write_per_mtok — Anthropic only: tokens written into the prompt cache
+  cache_read_per_mtok  — tokens served from cache (cheaper rate)
+
+For OpenAI the API returns prompt_tokens (total) and
+prompt_tokens_details.cached_tokens (the subset from cache).
+Split them before calling cost():
+  uncached = prompt_tokens - cached_tokens   → input_tokens arg
+  cached   = cached_tokens                   → cache_read_tokens arg
 """
 
 from __future__ import annotations
@@ -21,9 +36,9 @@ class ModelPricing:
     model_id: str
     display_name: str
     provider: str
-    input_per_mtok: float        # USD / 1M input tokens
+    input_per_mtok: float        # USD / 1M input tokens  (uncached)
     output_per_mtok: float       # USD / 1M output tokens
-    cache_write_per_mtok: float  # USD / 1M cache-write tokens
+    cache_write_per_mtok: float  # USD / 1M cache-write tokens  (Anthropic only)
     cache_read_per_mtok: float   # USD / 1M cache-read tokens
 
     def cost(
@@ -34,39 +49,115 @@ class ModelPricing:
         cache_read_tokens: int = 0,
     ) -> float:
         return (
-            (input_tokens        / 1_000_000) * self.input_per_mtok
-            + (output_tokens     / 1_000_000) * self.output_per_mtok
-            + (cache_write_tokens/ 1_000_000) * self.cache_write_per_mtok
-            + (cache_read_tokens / 1_000_000) * self.cache_read_per_mtok
+            (input_tokens         / 1_000_000) * self.input_per_mtok
+            + (output_tokens      / 1_000_000) * self.output_per_mtok
+            + (cache_write_tokens / 1_000_000) * self.cache_write_per_mtok
+            + (cache_read_tokens  / 1_000_000) * self.cache_read_per_mtok
         )
 
 
 _REGISTRY: list[ModelPricing] = [
-    # ── Anthropic ──────────────────────────────────────────────────────────
-    ModelPricing("claude-sonnet-4-20250514", "Claude Sonnet 4",    "anthropic",  3.00, 15.00,  3.75, 0.30),
-    ModelPricing("claude-sonnet-4-6",        "Claude Sonnet 4.6",  "anthropic",  3.00, 15.00,  3.75, 0.30),
-    ModelPricing("claude-haiku-4-5-20251001","Claude Haiku 4.5",   "anthropic",  1.00,  5.00,  1.25, 0.10),
-    ModelPricing("claude-opus-4-6",          "Claude Opus 4.6",    "anthropic",  5.00, 25.00,  6.25, 0.50),
-    ModelPricing("claude-opus-4-7",          "Claude Opus 4.7",    "anthropic",  5.00, 25.00,  6.25, 0.50),
-    # ── OpenAI / Copilot ──────────────────────────────────────────────────
-    ModelPricing("gpt-4o",                   "GPT-4o",             "openai",     2.50, 10.00,  0.00, 0.00),
-    ModelPricing("gpt-4o-mini",              "GPT-4o Mini",        "openai",     0.15,  0.60,  0.00, 0.00),
+    # ── Anthropic ─────────────────────────────────────────────────────────────
+    # Source: https://docs.anthropic.com/en/docs/about-claude/models/overview
+    #         input / output / cache_write(5min) / cache_read
+    ModelPricing("claude-sonnet-4-20250514", "Claude Sonnet 4",   "anthropic",  3.00, 15.00, 3.75, 0.30),
+    ModelPricing("claude-sonnet-4-6",        "Claude Sonnet 4.6", "anthropic",  3.00, 15.00, 3.75, 0.30),
+    ModelPricing("claude-haiku-4-5-20251001","Claude Haiku 4.5",  "anthropic",  1.00,  5.00, 1.25, 0.10),
+    ModelPricing("claude-opus-4-6",          "Claude Opus 4.6",   "anthropic",  5.00, 25.00, 6.25, 0.50),
+    ModelPricing("claude-opus-4-7",          "Claude Opus 4.7",   "anthropic",  5.00, 25.00, 6.25, 0.50),
+
+    # ── OpenAI / Copilot ──────────────────────────────────────────────────────
+    # Source: https://openai.com/api/pricing  (standard tier, April 2026)
+    #         input / output / cache_write(n/a=0) / cache_read
+    # GPT-5.x flagship
+    ModelPricing("gpt-5.4",      "GPT-5.4",       "openai",  2.50, 15.00, 0.00, 0.250),
+    ModelPricing("gpt-5.4-mini", "GPT-5.4 Mini",  "openai",  0.75,  4.50, 0.00, 0.075),
+    ModelPricing("gpt-5.4-nano", "GPT-5.4 Nano",  "openai",  0.20,  1.25, 0.00, 0.020),
+    ModelPricing("gpt-5.2",      "GPT-5.2",       "openai",  1.75, 14.00, 0.00, 0.175),
+    ModelPricing("gpt-5.1",      "GPT-5.1",       "openai",  1.25, 10.00, 0.00, 0.125),
+    ModelPricing("gpt-5",        "GPT-5",         "openai",  1.25, 10.00, 0.00, 0.125),
+    ModelPricing("gpt-5-mini",   "GPT-5 Mini",    "openai",  0.25,  2.00, 0.00, 0.025),
+    ModelPricing("gpt-5-nano",   "GPT-5 Nano",    "openai",  0.05,  0.40, 0.00, 0.005),
+    # GPT-4.1 family
+    ModelPricing("gpt-4.1",      "GPT-4.1",       "openai",  2.00,  8.00, 0.00, 0.500),
+    ModelPricing("gpt-4.1-mini", "GPT-4.1 Mini",  "openai",  0.40,  1.60, 0.00, 0.100),
+    ModelPricing("gpt-4.1-nano", "GPT-4.1 Nano",  "openai",  0.10,  0.40, 0.00, 0.025),
+    # GPT-4o family
+    ModelPricing("gpt-4o",       "GPT-4o",        "openai",  2.50, 10.00, 0.00, 1.250),
+    ModelPricing("gpt-4o-mini",  "GPT-4o Mini",   "openai",  0.15,  0.60, 0.00, 0.075),
+    # o-series reasoning models
+    ModelPricing("o1",           "o1",            "openai", 15.00, 60.00, 0.00, 7.500),
+    ModelPricing("o1-mini",      "o1 mini",       "openai",  1.10,  4.40, 0.00, 0.550),
+    ModelPricing("o3",           "o3",            "openai",  2.00,  8.00, 0.00, 0.500),
+    ModelPricing("o3-mini",      "o3 mini",       "openai",  1.10,  4.40, 0.00, 0.550),
+    ModelPricing("o4-mini",      "o4 mini",       "openai",  1.10,  4.40, 0.00, 0.275),
+
+    # ── xAI ───────────────────────────────────────────────────────────────────
+    # Source: https://docs.x.ai/docs/models
+    ModelPricing("grok-4",       "Grok 4",        "xai",     2.00,  6.00, 0.00, 0.00),
+    ModelPricing("grok-4-fast",  "Grok 4 Fast",   "xai",     0.20,  0.50, 0.00, 0.00),
+    ModelPricing("grok-3",       "Grok 3",        "xai",     3.00, 15.00, 0.00, 0.00),
+    ModelPricing("grok-3-mini",  "Grok 3 Mini",   "xai",     0.30,  0.50, 0.00, 0.00),
+
+    # ── DeepSeek ──────────────────────────────────────────────────────────────
+    # Source: https://api-docs.deepseek.com/quick_start/pricing
+    #         cache_read = cache-hit input rate ($0.028), cache_write not charged separately
+    ModelPricing("deepseek-chat",      "DeepSeek V3",       "deepseek", 0.28, 0.42, 0.00, 0.028),
+    ModelPricing("deepseek-reasoner",  "DeepSeek R1",       "deepseek", 0.28, 0.42, 0.00, 0.028),
+    ModelPricing("deepseek-r1",        "DeepSeek R1 (alt)", "deepseek", 0.28, 0.42, 0.00, 0.028),
+
+    # ── Local / Ollama / LM Studio ────────────────────────────────────────────
+    ModelPricing("llama3.2",     "Llama 3.2",     "ollama",    0.00,  0.00, 0.00, 0.00),
+    ModelPricing("llama3.1",     "Llama 3.1",     "ollama",    0.00,  0.00, 0.00, 0.00),
+    ModelPricing("local-model",  "Local Model",   "lm-studio", 0.00,  0.00, 0.00, 0.00),
 ]
 
 _by_id: dict[str, ModelPricing] = {p.model_id: p for p in _REGISTRY}
 
 
 def get_pricing(model_id: str) -> Optional[ModelPricing]:
-    """Exact match first, then longest substring match."""
+    """
+    Exact match → substring match → family fallback.
+
+    The family fallback ensures unknown/future model names (e.g. gpt-6, claude-sonnet-5)
+    always return a pricing estimate rather than None, so cost never shows as $0.000000
+    simply because a new model name isn't in the registry yet.
+    """
     key = model_id.lower().strip()
     if key in _by_id:
         return _by_id[key]
-    # Substring: "claude-sonnet-4-20250514" ⊇ "sonnet-4"
+
+    # Substring: "claude-sonnet-4-20250514" ⊇ "sonnet-4", "gpt-5" ⊆ "gpt-5.4"
     for p in _REGISTRY:
         if key in p.model_id or p.model_id in key:
             return p
+
+    # Family fallback — same-family estimate for unknown/future model versions
+    if "claude" in key:
+        return _by_id.get("claude-sonnet-4-6")
+    if "gpt-5" in key:
+        return _by_id.get("gpt-5")
+    if "gpt-4.1" in key:
+        return _by_id.get("gpt-4.1")
+    if any(k in key for k in ("gpt-4o", "gpt-4", "gpt")) or key in ("gpt",):
+        return _by_id.get("gpt-4o")
+    if key.startswith("o1") or "/o1" in key or key == "o1":
+        return _by_id.get("o1")
+    if key.startswith("o3") or "/o3" in key or key == "o3":
+        return _by_id.get("o3")
+    if key.startswith("o4") or "/o4" in key or key == "o4":
+        return _by_id.get("o4-mini")
+    if "grok-4" in key:
+        return _by_id.get("grok-4")
+    if "grok" in key:
+        return _by_id.get("grok-3")
+    if "deepseek" in key:
+        return _by_id.get("deepseek-chat")
+    if any(k in key for k in ("llama", "mistral", "qwen", "gemma", "phi")):
+        return _by_id.get("local-model")
+
     return None
 
 
 def list_models() -> list[ModelPricing]:
-    return list(_REGISTRY)
+    return list(_REGISTRY)