scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-06 02:48:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 103

Merge pull request #37763 from BerriAI/litellm_fix_cognition_swe_1_7_pricing

Browse source 
fix(cognition): price swe-1.7 at the standard tier, add swe-1.7-lightning
This commit is contained in:
Mateo Wang 2026-08-20 21:01:11 -07:00 • committed by GitHub
commit ecd18c5a82
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 72 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
litellm/model_prices_and_context_window_backup.json
View file

@ -49489,6 +49489,16 @@
"source": "https://docs.devin.ai/windsurf/plugins/cascade/models"
},
"cognition/swe-1.7": {
"input_cost_per_token": 5e-07,
"output_cost_per_token": 2.5e-06,
"cache_read_input_token_cost": 2e-07,
"litellm_provider": "cognition",
"mode": "chat",
"supports_function_calling": true,
"supports_prompt_caching": true,
"source": "https://docs.devin.ai/desktop/models"
},
"cognition/swe-1.7-lightning": {
"input_cost_per_token": 2.5e-06,
"output_cost_per_token": 1.25e-05,
"cache_read_input_token_cost": 1e-06,
@ -49496,7 +49506,7 @@
"mode": "chat",
"supports_function_calling": true,
"supports_prompt_caching": true,
"source": "https://docs.devin.ai/windsurf/plugins/cascade/models"
"source": "https://docs.devin.ai/desktop/models"
},
"pinstripes/ps/glm-4.5-air": {
"max_tokens": 128000,

12
model_prices_and_context_window.json
View file

@ -49489,6 +49489,16 @@
"source": "https://docs.devin.ai/windsurf/plugins/cascade/models"
},
"cognition/swe-1.7": {
"input_cost_per_token": 5e-07,
"output_cost_per_token": 2.5e-06,
"cache_read_input_token_cost": 2e-07,
"litellm_provider": "cognition",
"mode": "chat",
"supports_function_calling": true,
"supports_prompt_caching": true,
"source": "https://docs.devin.ai/desktop/models"
},
"cognition/swe-1.7-lightning": {
"input_cost_per_token": 2.5e-06,
"output_cost_per_token": 1.25e-05,
"cache_read_input_token_cost": 1e-06,
@ -49496,7 +49506,7 @@
"mode": "chat",
"supports_function_calling": true,
"supports_prompt_caching": true,
"source": "https://docs.devin.ai/windsurf/plugins/cascade/models"
"source": "https://docs.devin.ai/desktop/models"
},
"pinstripes/ps/glm-4.5-air": {
"max_tokens": 128000,

58
tests/test_litellm/llms/openai_like/test_cognition_provider.py
View file

@ -111,33 +111,51 @@ class TestCognitionProviderIdentity:
class TestCognitionCostTracking:
@pytest.mark.parametrize(
"model, input_cost, output_cost",
"model, input_cost, output_cost, cache_read_cost",
[
("cognition/swe-1.6", 5e-07, 2.5e-06),
("cognition/swe-1.7", 2.5e-06, 1.25e-05),
("cognition/swe-1.6", 5e-07, 2.5e-06, 2e-07),
("cognition/swe-1.7", 5e-07, 2.5e-06, 2e-07),
("cognition/swe-1.7-lightning", 2.5e-06, 1.25e-05, 1e-06),
],
)
def test_cost_map_entries(self, model: str, input_cost: float, output_cost: float):
def test_cost_map_entries(self, model: str, input_cost: float, output_cost: float, cache_read_cost: float):
info = litellm.get_model_info(model=model)
assert info["litellm_provider"] == "cognition"
assert info["mode"] == "chat"
assert info["input_cost_per_token"] == input_cost
assert info["output_cost_per_token"] == output_cost
assert info["cache_read_input_token_cost"] == cache_read_cost
def test_cost_differs_from_openai_pricing(self):
@pytest.mark.parametrize(
"model, expected_prompt_cost, expected_completion_cost",
[
("cognition/swe-1.7", 0.5, 2.5),
("cognition/swe-1.7-lightning", 2.5, 12.5),
],
)
def test_cost_differs_from_openai_pricing(
self, model: str, expected_prompt_cost: float, expected_completion_cost: float
):
"""A cognition-prefixed model must never be priced off an OpenAI cost entry."""
from litellm.cost_calculator import cost_per_token
prompt_cost, completion_cost = cost_per_token(
model="cognition/swe-1.7",
model=model,
prompt_tokens=1_000_000,
completion_tokens=1_000_000,
custom_llm_provider="cognition",
)
assert prompt_cost == pytest.approx(2.5)
assert completion_cost == pytest.approx(12.5)
assert prompt_cost == pytest.approx(expected_prompt_cost)
assert completion_cost == pytest.approx(expected_completion_cost)
def test_lightning_is_five_times_the_standard_tier(self):
standard = litellm.get_model_info(model="cognition/swe-1.7")
lightning = litellm.get_model_info(model="cognition/swe-1.7-lightning")
assert lightning["input_cost_per_token"] == pytest.approx(standard["input_cost_per_token"] * 5)
assert lightning["output_cost_per_token"] == pytest.approx(standard["output_cost_per_token"] * 5)
def test_supported_endpoints_matrix(self):
matrix = json.loads((Path(litellm.__file__).parent / "provider_endpoints_support_backup.json").read_text())
@ -170,6 +188,30 @@ class TestCognitionRouting:
mock_response="hello from swe",
)
usage = response.usage
expected = usage.prompt_tokens * 5e-07 + usage.completion_tokens * 2.5e-06
assert response._hidden_params["response_cost"] == pytest.approx(expected)
@pytest.mark.asyncio
async def test_router_spend_uses_the_lightning_entry_for_lightning(self):
"""The Lightning tier is its own model, costed off its own entry."""
from litellm import Router
router = Router(
model_list=[
{
"model_name": "swe-lightning",
"litellm_params": {"model": "cognition/swe-1.7-lightning", "api_key": "sk-test"},
}
]
)
response = await router.acompletion(
model="swe-lightning",
messages=[{"role": "user", "content": "hi"}],
mock_response="hello from swe lightning",
)
usage = response.usage
expected = usage.prompt_tokens * 2.5e-06 + usage.completion_tokens * 1.25e-05
assert response._hidden_params["response_cost"] == pytest.approx(expected)