mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-08 22:21:35 +00:00
Merge pull request #39170 from BerriAI/litellm_registry_audit_2026_09_01
fix(models): registry audit 2026-09-01: openai realtime and long-context tiers, mistral aliases, voyage, xai, fireworks, together, scaleway, azure ai, govcloud, azure gov, cloudflare whisper, deprecation dates
This commit is contained in:
Mateo Wang
GitHub
commit
4049a075bd
7 changed files with 1912 additions and 125 deletions
File diff suppressed because it is too large
Load diff
File diff suppressed because it is too large
Load diff
|
|
@ -1522,7 +1522,7 @@ def test_gpt_5_6_alias_prices_match_sol(local_model_cost_map):
|
|||
sol = litellm.model_cost["gpt-5.6-sol"]
|
||||
|
||||
cost_fields = sorted(field for field in sol if "cost" in field)
|
||||
assert len(cost_fields) == 23
|
||||
assert len(cost_fields) == 27
|
||||
|
||||
for field in cost_fields:
|
||||
assert alias.get(field) == sol.get(field), field
|
||||
|
|
@ -4039,8 +4039,8 @@ def test_fast_service_tier_matches_priority_above_the_context_threshold(_local_m
|
|||
)
|
||||
|
||||
assert fast == priority
|
||||
assert fast[0] == pytest.approx(300_000 * 8e-06, rel=1e-9)
|
||||
assert fast[1] == pytest.approx(1_000 * 3e-05, rel=1e-9)
|
||||
assert fast[0] == pytest.approx(300_000 * 1.6e-05, rel=1e-9)
|
||||
assert fast[1] == pytest.approx(1_000 * 6e-05, rel=1e-9)
|
||||
|
||||
|
||||
def test_priority_reasoning_tokens_bill_at_the_priority_output_rate(_local_model_cost_map):
|
||||
|
|
|
|||
|
|
@ -26,9 +26,7 @@ import pytest
|
|||
|
||||
@pytest.fixture(scope="module")
|
||||
def model_data():
|
||||
json_path = os.path.join(
|
||||
os.path.dirname(__file__), "../../model_prices_and_context_window.json"
|
||||
)
|
||||
json_path = os.path.join(os.path.dirname(__file__), "../../model_prices_and_context_window.json")
|
||||
with open(json_path) as f:
|
||||
return json.load(f)
|
||||
|
||||
|
|
@ -51,21 +49,14 @@ def test_usgov_sonnet_4_5_pricing(model_data, model_key):
|
|||
info = model_data[model_key]
|
||||
|
||||
assert info["input_cost_per_token"] == 3.6e-06, (
|
||||
f"{model_key}: input_cost_per_token should be $3.60/MTok "
|
||||
f"(got {info['input_cost_per_token']})"
|
||||
f"{model_key}: input_cost_per_token should be $3.60/MTok (got {info['input_cost_per_token']})"
|
||||
)
|
||||
assert (
|
||||
info["output_cost_per_token"] == 1.8e-05
|
||||
), f"{model_key}: output_cost_per_token should be $18.00/MTok"
|
||||
assert (
|
||||
info["cache_creation_input_token_cost"] == 4.5e-06
|
||||
), f"{model_key}: 5m cache write should be $4.50/MTok"
|
||||
assert (
|
||||
info["cache_creation_input_token_cost_above_1hr"] == 7.2e-06
|
||||
), f"{model_key}: 1h cache write should be $7.20/MTok"
|
||||
assert (
|
||||
info["cache_read_input_token_cost"] == 3.6e-07
|
||||
), f"{model_key}: cache read should be $0.36/MTok"
|
||||
assert info["output_cost_per_token"] == 1.8e-05, f"{model_key}: output_cost_per_token should be $18.00/MTok"
|
||||
assert info["cache_creation_input_token_cost"] == 4.5e-06, f"{model_key}: 5m cache write should be $4.50/MTok"
|
||||
assert info["cache_creation_input_token_cost_above_1hr"] == 7.2e-06, (
|
||||
f"{model_key}: 1h cache write should be $7.20/MTok"
|
||||
)
|
||||
assert info["cache_read_input_token_cost"] == 3.6e-07, f"{model_key}: cache read should be $0.36/MTok"
|
||||
|
||||
|
||||
def test_usgov_carries_20_percent_premium_over_global(model_data):
|
||||
|
|
@ -84,9 +75,7 @@ def test_usgov_carries_20_percent_premium_over_global(model_data):
|
|||
"cache_read_input_token_cost",
|
||||
):
|
||||
ratio = usgov_info[field] / global_info[field]
|
||||
assert (
|
||||
abs(ratio - 1.2) < 1e-9
|
||||
), f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
|
||||
assert abs(ratio - 1.2) < 1e-9, f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
|
||||
|
||||
|
||||
# The us-gov.anthropic.* cross-region inference profile is the only us-gov
|
||||
|
|
@ -112,9 +101,7 @@ def test_usgov_cross_region_above_200k_carries_gov_premium(model_data, field, ex
|
|||
"""
|
||||
info = model_data[USGOV_CROSS_REGION_KEY]
|
||||
assert field in info, f"{USGOV_CROSS_REGION_KEY}: missing field {field}"
|
||||
assert (
|
||||
info[field] == expected
|
||||
), f"{USGOV_CROSS_REGION_KEY}: {field} should be {expected} (got {info[field]})"
|
||||
assert info[field] == expected, f"{USGOV_CROSS_REGION_KEY}: {field} should be {expected} (got {info[field]})"
|
||||
|
||||
|
||||
def test_usgov_cross_region_above_200k_ratio_to_global(model_data):
|
||||
|
|
@ -127,6 +114,176 @@ def test_usgov_cross_region_above_200k_ratio_to_global(model_data):
|
|||
usgov_info = model_data[USGOV_CROSS_REGION_KEY]
|
||||
for field in EXPECTED_USGOV_ABOVE_200K:
|
||||
ratio = usgov_info[field] / global_info[field]
|
||||
assert (
|
||||
abs(ratio - 1.2) < 1e-9
|
||||
), f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
|
||||
assert abs(ratio - 1.2) < 1e-9, f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
|
||||
|
||||
|
||||
CLAUDE_GOV_EXPECTED = {
|
||||
"anthropic.claude-sonnet-5": {
|
||||
"input_cost_per_token": 2.4e-06,
|
||||
"output_cost_per_token": 1.2e-05,
|
||||
"cache_creation_input_token_cost": 3e-06,
|
||||
"cache_creation_input_token_cost_above_1hr": 4.8e-06,
|
||||
"cache_read_input_token_cost": 2.4e-07,
|
||||
},
|
||||
"anthropic.claude-opus-4-8": {
|
||||
"input_cost_per_token": 6e-06,
|
||||
"output_cost_per_token": 3e-05,
|
||||
"cache_creation_input_token_cost": 7.5e-06,
|
||||
"cache_creation_input_token_cost_above_1hr": 1.2e-05,
|
||||
"cache_read_input_token_cost": 6e-07,
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
USGOV_CLAUDE_KEY_TEMPLATES = {
|
||||
"bedrock/us-gov-east-1/{base_key}": "bedrock",
|
||||
"bedrock/us-gov-west-1/{base_key}": "bedrock",
|
||||
"us-gov.{base_key}": "bedrock_converse",
|
||||
}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("base_key", CLAUDE_GOV_EXPECTED)
|
||||
@pytest.mark.parametrize("key_template,expected_provider", USGOV_CLAUDE_KEY_TEMPLATES.items())
|
||||
def test_usgov_claude_sonnet5_opus48_pricing(model_data, key_template, expected_provider, base_key):
|
||||
"""Sonnet 5 and Opus 4.8 gov entries, both in-region keys and the us-gov.
|
||||
geo inference profile the model cards list for GovCloud, must match the
|
||||
rates AWS publishes on the Bedrock pricing page (1.2x global).
|
||||
"""
|
||||
gov_key = key_template.format(base_key=base_key)
|
||||
assert gov_key in model_data, f"Missing model entry: {gov_key}"
|
||||
info = model_data[gov_key]
|
||||
assert info["litellm_provider"] == expected_provider
|
||||
for field, expected in CLAUDE_GOV_EXPECTED[base_key].items():
|
||||
assert info[field] == expected, f"{gov_key}: {field} should be {expected} (got {info[field]})"
|
||||
ratio = info[field] / model_data[base_key][field]
|
||||
assert abs(ratio - 1.2) < 1e-9, f"{gov_key}: {field} gov/global ratio is {ratio}, expected 1.2"
|
||||
|
||||
|
||||
CONVERSE_GOV_EXPECTED = {
|
||||
"nvidia.nemotron-nano-3-30b": (7.2e-08, 2.88e-07),
|
||||
"nvidia.nemotron-nano-12b-v2": (2.4e-07, 7.2e-07),
|
||||
"nvidia.nemotron-super-3-120b": (1.8e-07, 7.8e-07),
|
||||
"openai.gpt-oss-20b-1:0": (8.4e-08, 3.6e-07),
|
||||
"openai.gpt-oss-120b-1:0": (1.8e-07, 7.2e-07),
|
||||
}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("base_key", CONVERSE_GOV_EXPECTED)
|
||||
@pytest.mark.parametrize("region", ["us-gov-east-1", "us-gov-west-1"])
|
||||
def test_usgov_converse_model_pricing(model_data, region, base_key):
|
||||
"""Nemotron and gpt-oss gov entries must match the AWS Bedrock offer file,
|
||||
which prices both GovCloud regions identically at 1.2x commercial.
|
||||
"""
|
||||
gov_key = f"bedrock/{region}/{base_key}"
|
||||
assert gov_key in model_data, f"Missing model entry: {gov_key}"
|
||||
info = model_data[gov_key]
|
||||
expected_input, expected_output = CONVERSE_GOV_EXPECTED[base_key]
|
||||
assert info["input_cost_per_token"] == expected_input
|
||||
assert info["output_cost_per_token"] == expected_output
|
||||
assert info["litellm_provider"] == "bedrock"
|
||||
base = model_data[base_key]
|
||||
assert abs(info["input_cost_per_token"] / base["input_cost_per_token"] - 1.2) < 1e-9
|
||||
assert abs(info["output_cost_per_token"] / base["output_cost_per_token"] - 1.2) < 1e-9
|
||||
|
||||
|
||||
def test_usgov_west_llama3_8b_output_price_fixed(model_data):
|
||||
"""The us-gov-west-1 llama3-8b entry carried the 70B output rate ($2.65/MTok);
|
||||
the AWS Bedrock offer file prices output at $0.60/MTok. AWS lists the model
|
||||
in us-gov-west-1 only, so there is no east entry to check.
|
||||
"""
|
||||
info = model_data["bedrock/us-gov-west-1/meta.llama3-8b-instruct-v1:0"]
|
||||
assert info["input_cost_per_token"] == 3e-07
|
||||
assert info["output_cost_per_token"] == 6e-07
|
||||
|
||||
|
||||
MANTLE_GOV_TIERED_EXPECTED = {
|
||||
"openai.gpt-5.6-luna": {
|
||||
"input_cost_per_token": 2.64e-07,
|
||||
"input_cost_per_token_above_272k_tokens": 5.28e-07,
|
||||
"cache_creation_input_token_cost": 3.3e-07,
|
||||
"cache_creation_input_token_cost_above_272k_tokens": 6.6e-07,
|
||||
"cache_read_input_token_cost": 2.64e-08,
|
||||
"cache_read_input_token_cost_above_272k_tokens": 5.28e-08,
|
||||
"output_cost_per_token": 1.584e-06,
|
||||
"output_cost_per_token_above_272k_tokens": 2.376e-06,
|
||||
},
|
||||
"openai.gpt-5.6-terra": {
|
||||
"input_cost_per_token": 2.64e-06,
|
||||
"input_cost_per_token_above_272k_tokens": 5.28e-06,
|
||||
"cache_creation_input_token_cost": 3.3e-06,
|
||||
"cache_creation_input_token_cost_above_272k_tokens": 6.6e-06,
|
||||
"cache_read_input_token_cost": 2.64e-07,
|
||||
"cache_read_input_token_cost_above_272k_tokens": 5.28e-07,
|
||||
"output_cost_per_token": 1.584e-05,
|
||||
"output_cost_per_token_above_272k_tokens": 2.376e-05,
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model", MANTLE_GOV_TIERED_EXPECTED)
|
||||
def test_usgov_west_mantle_terra_luna_pricing(model_data, model):
|
||||
"""Terra and Luna carry 1.2x commercial across every tier in the
|
||||
us-gov-west-1 offer file; the us-gov-east-1 offer file has no SKUs for them.
|
||||
"""
|
||||
gov_key = f"bedrock_mantle/us-gov-west-1/{model}"
|
||||
assert gov_key in model_data, f"Missing model entry: {gov_key}"
|
||||
info = model_data[gov_key]
|
||||
for field, expected in MANTLE_GOV_TIERED_EXPECTED[model].items():
|
||||
assert info[field] == expected, f"{gov_key}: {field} should be {expected} (got {info[field]})"
|
||||
assert info["litellm_provider"] == "bedrock_mantle"
|
||||
assert f"bedrock_mantle/us-gov-east-1/{model}" not in model_data
|
||||
|
||||
|
||||
@pytest.mark.parametrize("region", ["us-gov-east-1", "us-gov-west-1"])
|
||||
def test_usgov_mantle_gpt_5_4_pricing_has_no_long_context_tier(model_data, region):
|
||||
"""gpt-5.4 gov rates come from the offer file, which publishes only the
|
||||
standard tier in GovCloud: no long-context SKUs exist there, unlike commercial.
|
||||
"""
|
||||
gov_key = f"bedrock_mantle/{region}/openai.gpt-5.4"
|
||||
assert gov_key in model_data, f"Missing model entry: {gov_key}"
|
||||
info = model_data[gov_key]
|
||||
assert info["input_cost_per_token"] == 3.3e-06
|
||||
assert info["cache_read_input_token_cost"] == 3.3e-07
|
||||
assert info["output_cost_per_token"] == 1.98e-05
|
||||
assert not any(field.endswith("_above_272k_tokens") for field in info)
|
||||
|
||||
|
||||
def test_usgov_mantle_grok_4_3_west_only(model_data):
|
||||
"""grok-4.3 is priced in the us-gov-west-1 offer file only; the east offer
|
||||
file carries grok-4.6 instead.
|
||||
"""
|
||||
info = model_data["bedrock_mantle/us-gov-west-1/xai.grok-4.3"]
|
||||
assert info["input_cost_per_token"] == 1.5e-06
|
||||
assert info["output_cost_per_token"] == 3e-06
|
||||
assert info["cache_read_input_token_cost"] == 2.4e-07
|
||||
assert "bedrock_mantle/us-gov-east-1/xai.grok-4.3" not in model_data
|
||||
|
||||
|
||||
AZURE_GOV_EXPECTED = {
|
||||
"azure/us-gov/gpt-5.1": {
|
||||
"input_cost_per_token": 1.71875e-06,
|
||||
"cache_read_input_token_cost": 1.71875e-07,
|
||||
"output_cost_per_token": 1.375e-05,
|
||||
},
|
||||
"azure/us-gov/o3-mini": {
|
||||
"input_cost_per_token": 1.513e-06,
|
||||
"cache_read_input_token_cost": 7.57e-07,
|
||||
"output_cost_per_token": 6.05e-06,
|
||||
},
|
||||
"azure/us-gov/text-embedding-3-large": {"input_cost_per_token": 1.63e-07},
|
||||
"azure/us-gov/text-embedding-3-small": {"input_cost_per_token": 2.5e-08},
|
||||
}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("gov_key", AZURE_GOV_EXPECTED)
|
||||
def test_azure_usgov_pricing(model_data, gov_key):
|
||||
"""Azure Government meters from the Azure retail prices API
|
||||
(usgovvirginia/usgovarizona, serviceName 'Foundry Models'). No Government
|
||||
retirement schedule is published, so these entries carry no deprecation_date.
|
||||
"""
|
||||
assert gov_key in model_data, f"Missing model entry: {gov_key}"
|
||||
info = model_data[gov_key]
|
||||
for field, expected in AZURE_GOV_EXPECTED[gov_key].items():
|
||||
assert info[field] == expected, f"{gov_key}: {field} should be {expected} (got {info[field]})"
|
||||
assert info["litellm_provider"] == "azure"
|
||||
assert "deprecation_date" not in info
|
||||
|
|
|
|||
|
|
@ -75,6 +75,22 @@ def test_additional_current_models_are_present():
|
|||
assert entry["output_cost_per_token"] > 0
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"key, published_price_per_audio_minute",
|
||||
[
|
||||
("cloudflare/@cf/openai/whisper", 0.00045),
|
||||
("cloudflare/@cf/openai/whisper-large-v3-turbo", 0.00051),
|
||||
],
|
||||
)
|
||||
def test_whisper_transcription_pricing_is_stored_per_second(key, published_price_per_audio_minute):
|
||||
entry = litellm.model_cost[key]
|
||||
assert entry["litellm_provider"] == "cloudflare"
|
||||
assert entry["mode"] == "audio_transcription"
|
||||
assert entry["supported_endpoints"] == ["/v1/audio/transcriptions"]
|
||||
assert entry["output_cost_per_second"] == 0.0
|
||||
assert entry["input_cost_per_second"] == pytest.approx(published_price_per_audio_minute / 60)
|
||||
|
||||
|
||||
def test_root_and_backup_have_identical_cloudflare_keys():
|
||||
if not os.path.exists(ROOT_MAP):
|
||||
pytest.skip("root cost map only ships in source checkouts")
|
||||
|
|
|
|||
|
|
@ -0,0 +1,156 @@
|
|||
import json
|
||||
from functools import lru_cache
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
|
||||
import litellm
|
||||
|
||||
REPO_ROOT = Path(__file__).parents[2]
|
||||
MAIN_PATH = REPO_ROOT / "model_prices_and_context_window.json"
|
||||
BACKUP_PATH = REPO_ROOT / "litellm" / "model_prices_and_context_window_backup.json"
|
||||
|
||||
FLEX_LONG_CONTEXT = {
|
||||
"gpt-5.4": {
|
||||
"input_cost_per_token_above_272k_tokens_flex": 2.5e-06,
|
||||
"output_cost_per_token_above_272k_tokens_flex": 1.125e-05,
|
||||
"cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07,
|
||||
},
|
||||
"gpt-5.4-pro": {
|
||||
"input_cost_per_token_above_272k_tokens_flex": 3e-05,
|
||||
"output_cost_per_token_above_272k_tokens_flex": 0.000135,
|
||||
},
|
||||
"gpt-5.5": {
|
||||
"input_cost_per_token_above_272k_tokens_flex": 5e-06,
|
||||
"output_cost_per_token_above_272k_tokens_flex": 2.25e-05,
|
||||
"cache_read_input_token_cost_above_272k_tokens_flex": 5e-07,
|
||||
},
|
||||
}
|
||||
|
||||
PRIORITY_LONG_CONTEXT = {
|
||||
"gpt-5.6": {
|
||||
"input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
|
||||
"output_cost_per_token_above_272k_tokens_priority": 6e-05,
|
||||
"cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
|
||||
"cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
|
||||
},
|
||||
"gpt-5.6-sol": {
|
||||
"input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
|
||||
"output_cost_per_token_above_272k_tokens_priority": 6e-05,
|
||||
"cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
|
||||
"cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
|
||||
},
|
||||
"gpt-5.6-terra": {
|
||||
"input_cost_per_token_above_272k_tokens_priority": 8e-06,
|
||||
"output_cost_per_token_above_272k_tokens_priority": 3.6e-05,
|
||||
"cache_read_input_token_cost_above_272k_tokens_priority": 8e-07,
|
||||
"cache_creation_input_token_cost_above_272k_tokens_priority": 1e-05,
|
||||
},
|
||||
"gpt-5.6-luna": {
|
||||
"input_cost_per_token_above_272k_tokens_priority": 8e-07,
|
||||
"output_cost_per_token_above_272k_tokens_priority": 3.6e-06,
|
||||
"cache_read_input_token_cost_above_272k_tokens_priority": 8e-08,
|
||||
"cache_creation_input_token_cost_above_272k_tokens_priority": 1e-06,
|
||||
},
|
||||
}
|
||||
|
||||
EXPECTED = {**FLEX_LONG_CONTEXT, **PRIORITY_LONG_CONTEXT}
|
||||
|
||||
NO_PUBLISHED_PRIORITY_LONG_CONTEXT = ("gpt-5.4", "gpt-5.5")
|
||||
|
||||
|
||||
@pytest.fixture(autouse=True)
|
||||
def _local_model_cost_map(monkeypatch: pytest.MonkeyPatch) -> None:
|
||||
monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
|
||||
monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url=""))
|
||||
|
||||
|
||||
@lru_cache(maxsize=2)
|
||||
def _load(path: Path) -> dict[str, dict[str, object]]:
|
||||
with open(path) as f:
|
||||
return json.load(f)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("path", [MAIN_PATH, BACKUP_PATH], ids=["main", "backup"])
|
||||
@pytest.mark.parametrize("model", sorted(EXPECTED))
|
||||
def test_service_tier_long_context_rates_are_published(model: str, path: Path) -> None:
|
||||
"""Each tier must carry its own above-272K rates, in both price files."""
|
||||
info = _load(path).get(model)
|
||||
assert info is not None, f"{model} not found in {path.name}"
|
||||
for key, expected in EXPECTED[model].items():
|
||||
assert info.get(key) == pytest.approx(expected), f"{model}.{key} is {info.get(key)!r}, expected {expected!r}"
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model", sorted(EXPECTED))
|
||||
def test_tier_long_context_rate_is_half_or_double_the_standard(model: str) -> None:
|
||||
"""Flex is half the standard long-context rate; priority is double it."""
|
||||
info = _load(MAIN_PATH)[model]
|
||||
tier = "flex" if model in FLEX_LONG_CONTEXT else "priority"
|
||||
ratio = 0.5 if tier == "flex" else 2.0
|
||||
for base in ("input_cost_per_token", "output_cost_per_token"):
|
||||
standard = info[f"{base}_above_272k_tokens"]
|
||||
tiered = info[f"{base}_above_272k_tokens_{tier}"]
|
||||
assert tiered == pytest.approx(standard * ratio), (
|
||||
f"{model}.{base}_above_272k_tokens_{tier} is {tiered!r}, "
|
||||
f"expected {ratio}x the standard long-context rate {standard!r}"
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model", NO_PUBLISHED_PRIORITY_LONG_CONTEXT)
|
||||
def test_no_priority_long_context_rates_where_openai_publishes_none(model: str) -> None:
|
||||
"""Guard against back-filling a rate OpenAI does not publish."""
|
||||
info = _load(MAIN_PATH)[model]
|
||||
assert "input_cost_per_token_above_272k_tokens_priority" not in info
|
||||
|
||||
|
||||
LONG_CONTEXT_PROMPT_TOKENS = 300_000
|
||||
COMPLETION_TOKENS = 1_000
|
||||
|
||||
TIERED_COST_CASES = [
|
||||
("gpt-5.4", "flex", 2.5e-06, 1.125e-05),
|
||||
("gpt-5.4-pro", "flex", 3e-05, 0.000135),
|
||||
("gpt-5.5", "flex", 5e-06, 2.25e-05),
|
||||
("gpt-5.6", "priority", 1.6e-05, 6e-05),
|
||||
("gpt-5.6-sol", "priority", 1.6e-05, 6e-05),
|
||||
("gpt-5.6-terra", "priority", 8e-06, 3.6e-05),
|
||||
("gpt-5.6-luna", "priority", 8e-07, 3.6e-06),
|
||||
]
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model,tier,input_rate,output_rate", TIERED_COST_CASES)
|
||||
def test_cost_per_token_bills_long_context_at_the_tier_rate(
|
||||
model: str, tier: str, input_rate: float, output_rate: float
|
||||
) -> None:
|
||||
"""A prompt over 272K on flex or priority must bill at that tier's long-context rate."""
|
||||
input_cost, output_cost = litellm.cost_per_token(
|
||||
model=model,
|
||||
prompt_tokens=LONG_CONTEXT_PROMPT_TOKENS,
|
||||
completion_tokens=COMPLETION_TOKENS,
|
||||
service_tier=tier,
|
||||
)
|
||||
assert input_cost == pytest.approx(LONG_CONTEXT_PROMPT_TOKENS * input_rate)
|
||||
assert output_cost == pytest.approx(COMPLETION_TOKENS * output_rate)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("model,tier,input_rate,output_rate", TIERED_COST_CASES)
|
||||
def test_cost_per_token_tier_differs_from_the_standard_long_context_cost(
|
||||
model: str, tier: str, input_rate: float, output_rate: float
|
||||
) -> None:
|
||||
"""Flex halves the standard long-context bill and priority doubles it."""
|
||||
ratio = 0.5 if tier == "flex" else 2.0
|
||||
standard = sum(
|
||||
litellm.cost_per_token(
|
||||
model=model,
|
||||
prompt_tokens=LONG_CONTEXT_PROMPT_TOKENS,
|
||||
completion_tokens=COMPLETION_TOKENS,
|
||||
)
|
||||
)
|
||||
tiered = sum(
|
||||
litellm.cost_per_token(
|
||||
model=model,
|
||||
prompt_tokens=LONG_CONTEXT_PROMPT_TOKENS,
|
||||
completion_tokens=COMPLETION_TOKENS,
|
||||
service_tier=tier,
|
||||
)
|
||||
)
|
||||
assert tiered == pytest.approx(standard * ratio)
|
||||
|
|
@ -217,3 +217,17 @@ bedrock/us-east-1/zai.glm-5
|
|||
bedrock/us-west-2/zai.glm-5
|
||||
bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0
|
||||
bedrock/us-gov-west-1/anthropic.claude-haiku-4-5-20251001-v1:0
|
||||
bedrock/us-gov-west-1/nvidia.nemotron-nano-3-30b
|
||||
bedrock/us-gov-west-1/nvidia.nemotron-nano-12b-v2
|
||||
bedrock/us-gov-west-1/nvidia.nemotron-super-3-120b
|
||||
bedrock/us-gov-west-1/openai.gpt-oss-20b-1:0
|
||||
bedrock/us-gov-west-1/openai.gpt-oss-120b-1:0
|
||||
bedrock/us-gov-west-1/anthropic.claude-sonnet-5
|
||||
bedrock/us-gov-west-1/anthropic.claude-opus-4-8
|
||||
bedrock/us-gov-east-1/nvidia.nemotron-nano-3-30b
|
||||
bedrock/us-gov-east-1/nvidia.nemotron-nano-12b-v2
|
||||
bedrock/us-gov-east-1/nvidia.nemotron-super-3-120b
|
||||
bedrock/us-gov-east-1/openai.gpt-oss-20b-1:0
|
||||
bedrock/us-gov-east-1/openai.gpt-oss-120b-1:0
|
||||
bedrock/us-gov-east-1/anthropic.claude-sonnet-5
|
||||
bedrock/us-gov-east-1/anthropic.claude-opus-4-8
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue