feat(pricing): refresh Fireworks snapshot to 2026-07, cover full serverless catalog + cached picker pricing

- Refresh _OFFICIAL_DOCS_PRICING fireworks entries against current
  docs.fireworks.ai/serverless/pricing: qwen3p6-plus is gone (replaced
  by qwen3p7-plus); add glm-5p2/5p1, kimi-k2p7-code, deepseek-v4-flash,
  minimax-m3/m2p7, gpt-oss-120b/20b, and the routers/*-fast tiers with
  their distinct higher rates.
- Picker pricing via get_pricing_for_provider('fireworks'): pure dict
  transform over the shared models.dev in-memory/disk cache (1h TTL) +
  _pricing_cache memoization — no new network call on the picker path.
- Wire pricing display into the generic api-key-provider setup flow so
  Fireworks model pickers show $/M columns like OpenRouter/Nous do.
- Invariant tests: plugin fallback_models all priced, fast tiers price
  higher than standard, every row carries cache_read < input.

hermes_cli/models.py

@@ -1654,6 +1654,8 @@ def get_pricing_for_provider(provider: str, *, force_refresh: bool = False) -> d
         return _fetch_novita_pricing(force_refresh=force_refresh)
     if normalized == "deepinfra":
         return _fetch_deepinfra_pricing(force_refresh=force_refresh)
+    if normalized == "fireworks":
+        return _fireworks_pricing_from_models_dev(force_refresh=force_refresh)
     if normalized == "nous":
         api_key, base_url = _resolve_nous_pricing_credentials()
         if base_url:
@@ -1670,6 +1672,55 @@ def get_pricing_for_provider(provider: str, *, force_refresh: bool = False) -> d
     return {}
 
 
+def _fireworks_pricing_from_models_dev(
+    *,
+    force_refresh: bool = False,
+) -> dict[str, dict[str, str]]:
+    """Derive Fireworks picker pricing from the models.dev registry cache.
+
+    No dedicated network fetch: ``fetch_models_dev()`` already maintains an
+    in-memory + disk cache (1h TTL) that every picker surface shares, so this
+    is a pure dict transform on the picker path — no added latency and no
+    per-render network call. Results are additionally memoized in
+    ``_pricing_cache`` so repeated menu renders within a process are free.
+
+    models.dev publishes Fireworks costs in USD per 1M tokens; the shared
+    pricing formatter expects per-token strings, so divide by 1M.
+    """
+    cache_key = "models.dev/fireworks"
+    if not force_refresh and cache_key in _pricing_cache:
+        return _pricing_cache[cache_key]
+
+    result: dict[str, dict[str, str]] = {}
+    try:
+        from agent.models_dev import _get_provider_models
+
+        models = _get_provider_models("fireworks") or {}
+        for mid, entry in models.items():
+            if not isinstance(entry, dict):
+                continue
+            cost = entry.get("cost")
+            if not isinstance(cost, dict):
+                continue
+            inp = cost.get("input")
+            out = cost.get("output")
+            if inp is None and out is None:
+                continue
+            row: dict[str, str] = {
+                "prompt": str(float(inp or 0) / 1_000_000),
+                "completion": str(float(out or 0) / 1_000_000),
+            }
+            cache_read = cost.get("cache_read")
+            if cache_read:
+                row["input_cache_read"] = str(float(cache_read) / 1_000_000)
+            result[str(mid)] = row
+    except Exception:
+        result = {}
+
+    _pricing_cache[cache_key] = result
+    return result
+
+
 def _fetch_novita_pricing(
     timeout: float = 8.0,
     *,