mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-06-24 10:52:21 +00:00
e0498bd305
Bedrock Claude routes through the AnthropicBedrock SDK and injects cache_control, so cached tokens are always reported — but the pricing table had no cache cost fields for any Bedrock model, so /usage showed "cost unknown" on every cached session. Also, cross-region inference profiles (us./global./eu. prefixes) never matched the bare pricing keys. - Add cache_read/cache_write rates to the four Bedrock Claude rows (read 0.1x input, write 1.25x input per the Bedrock pricing page). - Normalize the cross-region prefix in the Bedrock pricing lookup, mirroring is_anthropic_bedrock_model's prefix list. Closes #50295.
324 lines
11 KiB
Python
324 lines
11 KiB
Python
from types import SimpleNamespace
|
||
|
||
from agent.usage_pricing import (
|
||
CanonicalUsage,
|
||
estimate_usage_cost,
|
||
get_pricing_entry,
|
||
normalize_usage,
|
||
)
|
||
|
||
|
||
def test_normalize_usage_anthropic_keeps_cache_buckets_separate():
|
||
usage = SimpleNamespace(
|
||
input_tokens=1000,
|
||
output_tokens=500,
|
||
cache_read_input_tokens=2000,
|
||
cache_creation_input_tokens=400,
|
||
)
|
||
|
||
normalized = normalize_usage(usage, provider="anthropic", api_mode="anthropic_messages")
|
||
|
||
assert normalized.input_tokens == 1000
|
||
assert normalized.output_tokens == 500
|
||
assert normalized.cache_read_tokens == 2000
|
||
assert normalized.cache_write_tokens == 400
|
||
assert normalized.prompt_tokens == 3400
|
||
|
||
|
||
def test_normalize_usage_openai_subtracts_cached_prompt_tokens():
|
||
usage = SimpleNamespace(
|
||
prompt_tokens=3000,
|
||
completion_tokens=700,
|
||
prompt_tokens_details=SimpleNamespace(cached_tokens=1800),
|
||
)
|
||
|
||
normalized = normalize_usage(usage, provider="openai", api_mode="chat_completions")
|
||
|
||
assert normalized.input_tokens == 1200
|
||
assert normalized.cache_read_tokens == 1800
|
||
assert normalized.output_tokens == 700
|
||
|
||
|
||
def test_normalize_usage_openai_reads_top_level_anthropic_cache_fields():
|
||
"""Some OpenAI-compatible proxies (OpenRouter, Cline) expose
|
||
Anthropic-style cache token counts at the top level of the usage object when
|
||
routing Claude models, instead of nesting them in prompt_tokens_details.
|
||
|
||
Regression guard for the bug fixed in cline/cline#10266 — before this fix,
|
||
the chat-completions branch of normalize_usage() only read
|
||
prompt_tokens_details.cache_write_tokens and completely missed the
|
||
cache_creation_input_tokens case, so cache writes showed as 0 and reflected
|
||
inputTokens were overstated by the cache-write amount.
|
||
"""
|
||
usage = SimpleNamespace(
|
||
prompt_tokens=1000,
|
||
completion_tokens=200,
|
||
prompt_tokens_details=SimpleNamespace(cached_tokens=500),
|
||
cache_creation_input_tokens=300,
|
||
)
|
||
|
||
normalized = normalize_usage(usage, provider="openrouter", api_mode="chat_completions")
|
||
|
||
# Expected: cache read from prompt_tokens_details.cached_tokens (preferred),
|
||
# cache write from top-level cache_creation_input_tokens (fallback).
|
||
assert normalized.cache_read_tokens == 500
|
||
assert normalized.cache_write_tokens == 300
|
||
# input_tokens = prompt_total - cache_read - cache_write = 1000 - 500 - 300 = 200
|
||
assert normalized.input_tokens == 200
|
||
assert normalized.output_tokens == 200
|
||
|
||
|
||
def test_normalize_usage_openai_reads_top_level_cache_read_when_details_missing():
|
||
"""Some proxies expose only top-level Anthropic-style fields with no
|
||
prompt_tokens_details object. Regression guard for cline/cline#10266.
|
||
"""
|
||
usage = SimpleNamespace(
|
||
prompt_tokens=1000,
|
||
completion_tokens=200,
|
||
cache_read_input_tokens=500,
|
||
cache_creation_input_tokens=300,
|
||
)
|
||
|
||
normalized = normalize_usage(usage, provider="openrouter", api_mode="chat_completions")
|
||
|
||
assert normalized.cache_read_tokens == 500
|
||
assert normalized.cache_write_tokens == 300
|
||
assert normalized.input_tokens == 200
|
||
|
||
|
||
def test_normalize_usage_openai_prefers_prompt_tokens_details_over_top_level():
|
||
"""When both prompt_tokens_details and top-level Anthropic fields are
|
||
present, we prefer the OpenAI-standard nested fields. Top-level Anthropic
|
||
fields are only a fallback when the nested ones are absent/zero.
|
||
"""
|
||
usage = SimpleNamespace(
|
||
prompt_tokens=1000,
|
||
completion_tokens=200,
|
||
prompt_tokens_details=SimpleNamespace(cached_tokens=600, cache_write_tokens=150),
|
||
# Intentionally different values — proving we ignore these when details exist.
|
||
cache_read_input_tokens=999,
|
||
cache_creation_input_tokens=999,
|
||
)
|
||
|
||
normalized = normalize_usage(usage, provider="openrouter", api_mode="chat_completions")
|
||
|
||
assert normalized.cache_read_tokens == 600
|
||
assert normalized.cache_write_tokens == 150
|
||
|
||
|
||
def test_openrouter_models_api_pricing_is_converted_from_per_token_to_per_million(monkeypatch):
|
||
monkeypatch.setattr(
|
||
"agent.usage_pricing.fetch_model_metadata",
|
||
lambda: {
|
||
"anthropic/claude-opus-4.6": {
|
||
"pricing": {
|
||
"prompt": "0.000005",
|
||
"completion": "0.000025",
|
||
"input_cache_read": "0.0000005",
|
||
"input_cache_write": "0.00000625",
|
||
}
|
||
}
|
||
},
|
||
)
|
||
|
||
entry = get_pricing_entry(
|
||
"anthropic/claude-opus-4.6",
|
||
provider="openrouter",
|
||
base_url="https://openrouter.ai/api/v1",
|
||
)
|
||
|
||
assert float(entry.input_cost_per_million) == 5.0
|
||
assert float(entry.output_cost_per_million) == 25.0
|
||
assert float(entry.cache_read_cost_per_million) == 0.5
|
||
assert float(entry.cache_write_cost_per_million) == 6.25
|
||
|
||
|
||
def test_estimate_usage_cost_marks_subscription_routes_included():
|
||
result = estimate_usage_cost(
|
||
"gpt-5.3-codex",
|
||
CanonicalUsage(input_tokens=1000, output_tokens=500),
|
||
provider="openai-codex",
|
||
base_url="https://chatgpt.com/backend-api/codex",
|
||
)
|
||
|
||
assert result.status == "included"
|
||
assert float(result.amount_usd) == 0.0
|
||
|
||
|
||
def test_estimate_usage_cost_refuses_cache_pricing_without_official_cache_rate(monkeypatch):
|
||
monkeypatch.setattr(
|
||
"agent.usage_pricing.fetch_model_metadata",
|
||
lambda: {
|
||
"google/gemini-2.5-pro": {
|
||
"pricing": {
|
||
"prompt": "0.00000125",
|
||
"completion": "0.00001",
|
||
}
|
||
}
|
||
},
|
||
)
|
||
|
||
result = estimate_usage_cost(
|
||
"google/gemini-2.5-pro",
|
||
CanonicalUsage(input_tokens=1000, output_tokens=500, cache_read_tokens=100),
|
||
provider="openrouter",
|
||
base_url="https://openrouter.ai/api/v1",
|
||
)
|
||
|
||
assert result.status == "unknown"
|
||
|
||
|
||
def test_custom_endpoint_models_api_pricing_is_supported(monkeypatch):
|
||
monkeypatch.setattr(
|
||
"agent.usage_pricing.fetch_endpoint_model_metadata",
|
||
lambda base_url, api_key=None: {
|
||
"zai-org/GLM-5-TEE": {
|
||
"pricing": {
|
||
"prompt": "0.0000005",
|
||
"completion": "0.000002",
|
||
}
|
||
}
|
||
},
|
||
)
|
||
|
||
entry = get_pricing_entry(
|
||
"zai-org/GLM-5-TEE",
|
||
provider="custom",
|
||
base_url="https://llm.chutes.ai/v1",
|
||
api_key="test-key",
|
||
)
|
||
|
||
assert float(entry.input_cost_per_million) == 0.5
|
||
assert float(entry.output_cost_per_million) == 2.0
|
||
|
||
|
||
def test_nous_portal_pricing_preserves_vendor_prefixed_model_ids(monkeypatch):
|
||
seen = {}
|
||
|
||
def _fake_fetch_endpoint_model_metadata(base_url, api_key=None):
|
||
seen["base_url"] = base_url
|
||
return {
|
||
"openai/gpt-5.5-pro": {
|
||
"pricing": {
|
||
"prompt": "0.000025",
|
||
"completion": "0.000125",
|
||
}
|
||
}
|
||
}
|
||
|
||
monkeypatch.setattr(
|
||
"agent.usage_pricing.fetch_endpoint_model_metadata",
|
||
_fake_fetch_endpoint_model_metadata,
|
||
)
|
||
|
||
entry = get_pricing_entry("openai/gpt-5.5-pro", provider="nous")
|
||
|
||
assert seen["base_url"] == "https://inference-api.nousresearch.com/v1"
|
||
assert float(entry.input_cost_per_million) == 25.0
|
||
assert float(entry.output_cost_per_million) == 125.0
|
||
|
||
|
||
def test_deepseek_v4_pro_pricing_entry_exists():
|
||
"""Regression test: deepseek-v4-pro must have a pricing entry.
|
||
|
||
Before this fix, deepseek-v4-pro sessions showed as unknown cost
|
||
in hermes insights because the _OFFICIAL_DOCS_PRICING table had no
|
||
entry for that model. See #24218.
|
||
"""
|
||
entry = get_pricing_entry(
|
||
"deepseek-v4-pro",
|
||
provider="deepseek",
|
||
)
|
||
|
||
assert entry is not None
|
||
assert entry.input_cost_per_million is not None
|
||
assert entry.output_cost_per_million is not None
|
||
assert float(entry.input_cost_per_million) == 1.74
|
||
assert float(entry.output_cost_per_million) == 3.48
|
||
assert float(entry.cache_read_cost_per_million) == 0.0145
|
||
|
||
|
||
def test_deepseek_v4_pro_estimate_usage_cost():
|
||
"""Ensure deepseek-v4-pro sessions get a dollar estimate, not unknown."""
|
||
result = estimate_usage_cost(
|
||
"deepseek-v4-pro",
|
||
CanonicalUsage(input_tokens=1000000, output_tokens=500000),
|
||
provider="deepseek",
|
||
)
|
||
|
||
assert result.status == "estimated"
|
||
assert result.amount_usd is not None
|
||
# 1M input × $1.74/M + 500K output × $3.48/M = $1.74 + $1.74 = $3.48
|
||
assert float(result.amount_usd) == 3.48
|
||
|
||
|
||
def test_bedrock_claude_rows_all_carry_cache_pricing():
|
||
"""Invariant: every Bedrock Claude pricing row must carry cache-read AND
|
||
cache-write rates, otherwise a cached session prices as ``unknown``.
|
||
|
||
Bedrock Claude routes through the AnthropicBedrock SDK and injects
|
||
cache_control, so cached tokens are always reported — the pricing layer
|
||
must be able to value them. See #50295.
|
||
"""
|
||
from agent.usage_pricing import _OFFICIAL_DOCS_PRICING
|
||
|
||
claude_rows = [
|
||
(prov, model)
|
||
for (prov, model) in _OFFICIAL_DOCS_PRICING
|
||
if prov == "bedrock" and "claude" in model
|
||
]
|
||
assert claude_rows, "expected at least one bedrock Claude pricing row"
|
||
for key in claude_rows:
|
||
entry = _OFFICIAL_DOCS_PRICING[key]
|
||
assert entry.input_cost_per_million is not None, key
|
||
assert entry.cache_read_cost_per_million is not None, key
|
||
assert entry.cache_write_cost_per_million is not None, key
|
||
# Cache reads are cheaper than fresh input; cache writes cost more.
|
||
assert entry.cache_read_cost_per_million < entry.input_cost_per_million, key
|
||
assert entry.cache_write_cost_per_million > entry.input_cost_per_million, key
|
||
|
||
|
||
def test_bedrock_cross_region_profile_prefix_resolves_to_pricing():
|
||
"""Cross-region inference profiles (us./global./eu. prefixes) must resolve
|
||
to the same pricing entry as the bare foundation-model id. Without prefix
|
||
normalization, ``us.anthropic.claude-*`` sessions price as unknown.
|
||
"""
|
||
bedrock_url = "https://bedrock-runtime.us-east-1.amazonaws.com"
|
||
bare = get_pricing_entry(
|
||
"anthropic.claude-sonnet-4-5", provider="bedrock", base_url=bedrock_url
|
||
)
|
||
assert bare is not None
|
||
for prefix in ("us.", "global.", "eu."):
|
||
scoped = get_pricing_entry(
|
||
f"{prefix}anthropic.claude-sonnet-4-5",
|
||
provider="bedrock",
|
||
base_url=bedrock_url,
|
||
)
|
||
assert scoped is not None, prefix
|
||
assert scoped.input_cost_per_million == bare.input_cost_per_million
|
||
assert scoped.cache_read_cost_per_million == bare.cache_read_cost_per_million
|
||
|
||
|
||
def test_bedrock_claude_cached_session_estimates_cost_not_unknown():
|
||
"""A Bedrock Claude session with cache hits must produce a dollar estimate,
|
||
not ``unknown`` — the user-visible symptom in #50295.
|
||
"""
|
||
bedrock_url = "https://bedrock-runtime.us-east-1.amazonaws.com"
|
||
usage = SimpleNamespace(
|
||
input_tokens=55,
|
||
output_tokens=7113,
|
||
cache_read_input_tokens=1369379,
|
||
cache_creation_input_tokens=42135,
|
||
)
|
||
canonical = normalize_usage(usage, provider="bedrock", api_mode="anthropic_messages")
|
||
assert canonical.cache_read_tokens == 1369379
|
||
assert canonical.cache_write_tokens == 42135
|
||
|
||
result = estimate_usage_cost(
|
||
"us.anthropic.claude-opus-4-6",
|
||
canonical,
|
||
provider="bedrock",
|
||
base_url=bedrock_url,
|
||
)
|
||
assert result.status == "estimated"
|
||
assert result.amount_usd is not None
|