Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

fix(tasks): offer deepseek v4.1 flash fast instead of silently running opus (#120877)

Alessandro Pogliaghi864782de217f
diff --git a/packages/agent/packages/agent-contracts/src/model-catalog.generated.ts b/packages/agent/packages/agent-contracts/src/model-catalog.generated.ts
index c265de6483f1..9bbdd9f5c61c 100644
--- a/packages/agent/packages/agent-contracts/src/model-catalog.generated.ts
+++ b/packages/agent/packages/agent-contracts/src/model-catalog.generated.ts
@@ -208,6 +208,18 @@ export const MODELS: readonly CatalogModel[] = [
costMultiplier: "≈0.05×",
costSummary: "Input $0.13 · Output $0.26 per 1M tokens",
},
+ {
+ id: "deepseek-ai/deepseek-v4.1-flash-1017-fast",
+ runtimeAdapter: "claude",
+ reasoningEfforts: [],
+ label: "DeepSeek V4.1 Flash Fast",
+ cost: {
+ inputPerMtok: 0.6,
+ outputPerMtok: 2.4,
+ },
+ costMultiplier: "≈0.27×",
+ costSummary: "Input $0.60 · Output $2.40 per 1M tokens",
+ },
{
id: "claude-opus-4-5",
runtimeAdapter: "claude",
diff --git a/packages/agent/packages/agent/src/adapters/base-acp-agent.test.ts b/packages/agent/packages/agent/src/adapters/base-acp-agent.test.ts
index 19d31d964237..0424e7e5fcd9 100644
--- a/packages/agent/packages/agent/src/adapters/base-acp-agent.test.ts
+++ b/packages/agent/packages/agent/src/adapters/base-acp-agent.test.ts
@@ -220,6 +220,7 @@ describe("BaseAcpAgent", () => {
it.each([
"moonshotai/kimi-k3",
"deepseek-ai/deepseek-v4-flash-0731",
+ "deepseek-ai/deepseek-v4.1-flash-1017-fast",
"zai-org/glm-5.3",
"zai-org/glm-5.3-flash",
])(
@@ -237,6 +238,7 @@ describe("BaseAcpAgent", () => {
]);
const agent = new TestAcpAgent({} as AgentSideConnection);
+ const warn = vi.spyOn(agent.logger, "warn");
const result = await agent.getModelConfigOptions(
savedModel,
"https://gateway.us.posthog.com/posthog_code",
@@ -247,6 +249,13 @@ describe("BaseAcpAgent", () => {
expect(result.options).not.toContainEqual(
expect.objectContaining({ value: savedModel }),
);
+ expect(warn).toHaveBeenCalledWith(
+ expect.any(String),
+ expect.objectContaining({
+ requestedModel: savedModel,
+ fallbackModel: "claude-opus-4-8",
+ }),
+ );
},
);
});
diff --git a/packages/agent/packages/agent/src/adapters/base-acp-agent.ts b/packages/agent/packages/agent/src/adapters/base-acp-agent.ts
index dad7571df79a..8ba08511d61e 100644
--- a/packages/agent/packages/agent/src/adapters/base-acp-agent.ts
+++ b/packages/agent/packages/agent/src/adapters/base-acp-agent.ts
@@ -200,6 +200,13 @@ export abstract class BaseAcpAgent implements Agent {
options.find((option) => option.value === DEFAULT_GATEWAY_MODEL) ??
options.find((option) => option._meta === undefined);
if (fallbackOption) {
+ this.logger.warn(
+ "Requested model is not offered on Claude adapter; falling back to default model",
+ {
+ requestedModel: currentModelId,
+ fallbackModel: fallbackOption.value,
+ },
+ );
currentModelId = fallbackOption.value;
}
} else if (!isClaudeAdapterModelId(currentModelId)) {
diff --git a/packages/agent/packages/agent/src/adapters/claude/session/budget-guard.ts b/packages/agent/packages/agent/src/adapters/claude/session/budget-guard.ts
index ec28c42fbd2d..4123b727b97a 100644
--- a/packages/agent/packages/agent/src/adapters/claude/session/budget-guard.ts
+++ b/packages/agent/packages/agent/src/adapters/claude/session/budget-guard.ts
@@ -106,6 +106,10 @@ export const DEFAULT_MODEL_PRICES: readonly ModelPriceRule[] = [
match: /kimi/i,
price: { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 3 },
},
+ {
+ match: /deepseek-v4\.1-flash-1017-fast/i,
+ price: { input: 0.6, output: 2.4, cacheRead: 0.014, cacheWrite: 0.6 },
+ },
{
match: /deepseek/i,
price: { input: 0.13, output: 0.26, cacheRead: 0.028, cacheWrite: 0.13 },
diff --git a/products/tasks/backend/logic/services/desktop_gateway_token.py b/products/tasks/backend/logic/services/desktop_gateway_token.py
index f72dd2633973..8613b72e115b 100644
--- a/products/tasks/backend/logic/services/desktop_gateway_token.py
+++ b/products/tasks/backend/logic/services/desktop_gateway_token.py
@@ -23,7 +23,7 @@ from prometheus_client import Counter
from posthog.constants import AvailableFeature
from posthog.exceptions_capture import capture_exception
-from products.tasks.backend.logic.services.gateway_model_pin import DESKTOP_AGENT_MODELS, FREE_TIER_MODELS
+from products.tasks.backend.logic.services.gateway_model_pin import DESKTOP_AGENT_MODELS, FREE_TIER_MODELS, gateway_pin
if TYPE_CHECKING:
from posthog.models.organization import Organization
@@ -305,7 +305,7 @@ def mint_desktop_gateway_token(
"product": POSTHOG_CODE_PRODUCT,
"obo": str(team_id),
"user": user,
- "allowed_models": allowed_models,
+ "allowed_models": gateway_pin(allowed_models),
"limit_tier": limit_tier,
}
try:
diff --git a/products/tasks/backend/logic/services/gateway_model_pin.py b/products/tasks/backend/logic/services/gateway_model_pin.py
index 2d5b5468c59e..94bec0bcd28a 100644
--- a/products/tasks/backend/logic/services/gateway_model_pin.py
+++ b/products/tasks/backend/logic/services/gateway_model_pin.py
@@ -13,6 +13,7 @@ SDK_IMPLICIT_MODELS: tuple[str, ...] = ("claude-haiku-4-5", "claude-sonnet-4-5")
_GATEWAY_RESTRICTED_MODEL_PRODUCTS: dict[str, frozenset[str]] = {
"deepseek-ai/deepseek-v4-flash-0731": frozenset({"posthog_code", "review_hog"}),
+ "deepseek-ai/deepseek-v4.1-flash-1017-fast": frozenset({"posthog_code", "review_hog", "signals_scout"}),
"zai-org/glm-5.3": frozenset({"posthog_code", "review_hog"}),
"zai-org/glm-5.3-flash": frozenset({"posthog_code", "review_hog"}),
}
@@ -34,13 +35,20 @@ _DESKTOP_LEGACY_MODELS: tuple[str, ...] = ("gpt-5.4", "gpt-5.3-codex", "gpt-5.2"
# Go has no `@cf/` alias, so a run stored on `@cf/zai-org/glm-5.2` calls Go under the bare id.
_GO_SPELLINGS: tuple[str, ...] = ("zai-org/glm-5.2",)
+
+def gateway_pin(models: list[str]) -> list[str]:
+ return [model for model in models if not model.startswith("@cf/")]
+
+
# Every model a PostHog Desktop session or cloud run may call, slash ids included. Go picks the provider.
DESKTOP_AGENT_MODELS: list[str] = list(
dict.fromkeys([*SDK_IMPLICIT_MODELS, *(model.id for model in MODELS), *_DESKTOP_LEGACY_MODELS, *_GO_SPELLINGS])
)
# The Go gateway refuses a pin longer than this at mint.
MAX_GATEWAY_PIN_LENGTH = 32
-assert len(DESKTOP_AGENT_MODELS) <= MAX_GATEWAY_PIN_LENGTH, "the desktop model pin outgrew the gateway limit"
+assert len(gateway_pin(DESKTOP_AGENT_MODELS)) <= MAX_GATEWAY_PIN_LENGTH, (
+ "the desktop model pin outgrew the gateway limit"
+)
# The legacy gateway's free tier. GLM-5.2 is listed twice: older clients send `@cf/`, Go serves the bare id.
FREE_TIER_MODELS: list[str] = [
diff --git a/products/tasks/backend/model_catalog.py b/products/tasks/backend/model_catalog.py
index 4fdf6bc8e3a5..9a35688944f4 100644
--- a/products/tasks/backend/model_catalog.py
+++ b/products/tasks/backend/model_catalog.py
@@ -164,6 +164,7 @@ _GLM_COST = ModelCost(1.4, 4.4)
_GLM_FLASH_COST = ModelCost(0.15, 0.5)
_KIMI_COST = ModelCost(3, 15)
_DEEPSEEK_COST = ModelCost(0.13, 0.26)
+_DEEPSEEK_V4_1_FLASH_FAST_COST = ModelCost(0.6, 2.4)
_OPUS_COST = ModelCost(5, 25)
_OPUS_5_5_COST = ModelCost(4, 20)
_FABLE_COST = ModelCost(10, 50)
@@ -196,6 +197,13 @@ MODELS: tuple[CatalogModel, ...] = (
label="DeepSeek V4 Flash",
cost=_DEEPSEEK_COST,
),
+ CatalogModel(
+ "deepseek-ai/deepseek-v4.1-flash-1017-fast",
+ CLAUDE,
+ _NO_EFFORT,
+ label="DeepSeek V4.1 Flash Fast",
+ cost=_DEEPSEEK_V4_1_FLASH_FAST_COST,
+ ),
CatalogModel("claude-opus-4-5", CLAUDE, _STANDARD, cost=_OPUS_COST, retired=True),
CatalogModel("claude-opus-4-6", CLAUDE, _THROUGH_MAX, cost=_OPUS_COST, retired=True),
CatalogModel(
diff --git a/products/tasks/backend/temporal/process_task/ai_gateway_token.py b/products/tasks/backend/temporal/process_task/ai_gateway_token.py
index d074e7c9d798..f5acf831f4f5 100644
--- a/products/tasks/backend/temporal/process_task/ai_gateway_token.py
+++ b/products/tasks/backend/temporal/process_task/ai_gateway_token.py
@@ -34,6 +34,7 @@ from products.tasks.backend.logic.services.desktop_gateway_token import (
from products.tasks.backend.logic.services.gateway_model_pin import (
FREE_TIER_MODELS,
PRODUCT_ALLOWED_MODELS,
+ gateway_pin,
model_allowed_by_product_pin,
)
from products.tasks.backend.logic.services.run_actor import is_slack_interaction_state
@@ -386,7 +387,7 @@ def mint_scoped_token(
body["user"] = user
pin = allowed_models if allowed_models is not None else _PRODUCT_ALLOWED_MODELS.get(ai_product)
if pin:
- body["allowed_models"] = pin
+ body["allowed_models"] = gateway_pin(pin)
if limit_tier:
body["limit_tier"] = limit_tier
if capture_mode is not None:
diff --git a/products/tasks/backend/temporal/process_task/tests/test_ai_gateway_token.py b/products/tasks/backend/temporal/process_task/tests/test_ai_gateway_token.py
index 96af99ecbb71..ac262b04ab14 100644
--- a/products/tasks/backend/temporal/process_task/tests/test_ai_gateway_token.py
+++ b/products/tasks/backend/temporal/process_task/tests/test_ai_gateway_token.py
@@ -37,6 +37,7 @@ from products.tasks.backend.logic.services.gateway_model_pin import (
FREE_TIER_PIN_KEY,
MAX_GATEWAY_PIN_LENGTH,
SDK_IMPLICIT_MODELS,
+ gateway_pin,
pinned_run_allows_model,
)
from products.tasks.backend.logic.services.sandbox_config import MAX_SANDBOX_TTL_SECONDS
@@ -309,7 +310,12 @@ class TestMintScopedToken:
assert len(pin) == len(set(pin))
expected = {model for model in registry if not model.startswith("@cf/")} | set(SDK_IMPLICIT_MODELS)
if product != "review_hog":
- expected -= {"deepseek-ai/deepseek-v4-flash-0731", "zai-org/glm-5.3", "zai-org/glm-5.3-flash"}
+ expected -= {
+ "deepseek-ai/deepseek-v4-flash-0731",
+ "deepseek-ai/deepseek-v4.1-flash-1017-fast",
+ "zai-org/glm-5.3",
+ "zai-org/glm-5.3-flash",
+ }
assert set(pin) == expected
def test_limit_tier_is_sent_only_when_given(self, mint_settings):
@@ -1254,7 +1260,12 @@ class TestSlackAppMint:
def test_pin_covers_every_routed_catalog_model(self):
routed = {entry.id for entry in model_catalog.MODELS if not entry.id.startswith("@cf/")}
- reserved = {"deepseek-ai/deepseek-v4-flash-0731", "zai-org/glm-5.3", "zai-org/glm-5.3-flash"}
+ reserved = {
+ "deepseek-ai/deepseek-v4-flash-0731",
+ "deepseek-ai/deepseek-v4.1-flash-1017-fast",
+ "zai-org/glm-5.3",
+ "zai-org/glm-5.3-flash",
+ }
assert reserved <= set(_PRODUCT_ALLOWED_MODELS["review_hog"])
missing = routed - reserved - set(_PRODUCT_ALLOWED_MODELS["slack_app"])
assert not missing, f"catalog models absent from the slack_app pin: {sorted(missing)}"
@@ -1522,7 +1533,8 @@ class TestPosthogCodeSandboxMint:
def test_pin_holds_every_catalog_id_and_fits_the_gateway(self):
pin = _PRODUCT_ALLOWED_MODELS["posthog_code"]
assert pin == DESKTOP_AGENT_MODELS
- assert len(pin) == len(set(pin)) <= MAX_GATEWAY_PIN_LENGTH
+ assert len(pin) == len(set(pin))
+ assert len(gateway_pin(pin)) <= MAX_GATEWAY_PIN_LENGTH
assert {entry.id for entry in model_catalog.MODELS} <= set(pin)
assert set(SDK_IMPLICIT_MODELS) <= set(pin)
assert {"gpt-5.4", "gpt-5.3-codex", "gpt-5.2", "gpt-5-mini"} <= set(pin)
@@ -1608,8 +1620,8 @@ class TestPosthogCodeSandboxMint:
post.return_value = self._mint_response()
mint_scoped_token(ai_product="posthog_code", team_id=2)
mint_scoped_token(ai_product="posthog_code", team_id=2, allowed_models=FREE_TIER_MODELS)
- assert post.call_args_list[0].kwargs["json"]["allowed_models"] == DESKTOP_AGENT_MODELS
- assert post.call_args_list[1].kwargs["json"]["allowed_models"] == FREE_TIER_MODELS
+ assert post.call_args_list[0].kwargs["json"]["allowed_models"] == gateway_pin(DESKTOP_AGENT_MODELS)
+ assert post.call_args_list[1].kwargs["json"]["allowed_models"] == gateway_pin(FREE_TIER_MODELS)
@pytest.mark.django_db
@pytest.mark.parametrize(
diff --git a/products/tasks/backend/tests/__snapshots__/test_model_catalog/test_every_model_resolves_to_the_recorded_triple.json b/products/tasks/backend/tests/__snapshots__/test_model_catalog/test_every_model_resolves_to_the_recorded_triple.json
index f9cb2195affa..9f7c05a2e7ee 100644
--- a/products/tasks/backend/tests/__snapshots__/test_model_catalog/test_every_model_resolves_to_the_recorded_triple.json
+++ b/products/tasks/backend/tests/__snapshots__/test_model_catalog/test_every_model_resolves_to_the_recorded_triple.json
@@ -301,6 +301,13 @@
"supports_1m_context": false,
"supports_fast_mode": false
},
+ "deepseek-ai/deepseek-v4.1-flash-1017-fast": {
+ "cost_multiplier": "≈0.27×",
+ "offered": true,
+ "reasoning_efforts": [],
+ "supports_1m_context": false,
+ "supports_fast_mode": false
+ },
"moonshotai/kimi-k3": {
"cost_multiplier": "1.5×",
"offered": true,
diff --git a/products/tasks/backend/tests/test_desktop_gateway_token.py b/products/tasks/backend/tests/test_desktop_gateway_token.py
index 7ad57f945d82..548e3b47dc29 100644
--- a/products/tasks/backend/tests/test_desktop_gateway_token.py
+++ b/products/tasks/backend/tests/test_desktop_gateway_token.py
@@ -27,7 +27,7 @@ from products.tasks.backend.logic.services.desktop_gateway_token import (
posthog_code_billing_synced,
posthog_code_trusted,
)
-from products.tasks.backend.logic.services.gateway_model_pin import DESKTOP_AGENT_MODELS, FREE_TIER_MODELS
+from products.tasks.backend.logic.services.gateway_model_pin import DESKTOP_AGENT_MODELS, FREE_TIER_MODELS, gateway_pin
from ee.billing.billing_manager import OrganizationFundingStatus, PrepaidCreditState
@@ -317,7 +317,7 @@ class TestDesktopGatewayTokenMint(_GatewayTestBase):
"product": "posthog_code",
"obo": str(self.team.id),
"user": self.user.distinct_id,
- "allowed_models": FREE_TIER_MODELS,
+ "allowed_models": gateway_pin(FREE_TIER_MODELS),
"limit_tier": "standard",
}
assert response.json() == {
@@ -353,7 +353,7 @@ class TestDesktopGatewayTokenMint(_GatewayTestBase):
self.organization.save()
response, post = self._mint(go=_minted(allowed_models=DESKTOP_AGENT_MODELS))
assert response.status_code == status.HTTP_201_CREATED
- assert post.call_args.kwargs["json"]["allowed_models"] == DESKTOP_AGENT_MODELS
+ assert post.call_args.kwargs["json"]["allowed_models"] == gateway_pin(DESKTOP_AGENT_MODELS)
assert response.json()["plan"] == "paid"
assert response.json()["allowed_models"] == DESKTOP_AGENT_MODELS
diff --git a/products/tasks/frontend/modelCatalog.generated.ts b/products/tasks/frontend/modelCatalog.generated.ts
index 47582bb0bfd7..24fbd77b9980 100644
--- a/products/tasks/frontend/modelCatalog.generated.ts
+++ b/products/tasks/frontend/modelCatalog.generated.ts
@@ -195,6 +195,18 @@ export const MODELS: readonly CatalogModel[] = [
costMultiplier: '≈0.05×',
costSummary: 'Input $0.13 · Output $0.26 per 1M tokens',
},
+ {
+ id: 'deepseek-ai/deepseek-v4.1-flash-1017-fast',
+ runtimeAdapter: 'claude',
+ reasoningEfforts: [],
+ label: 'DeepSeek V4.1 Flash Fast',
+ cost: {
+ inputPerMtok: 0.6,
+ outputPerMtok: 2.4,
+ },
+ costMultiplier: '≈0.27×',
+ costSummary: 'Input $0.60 · Output $2.40 per 1M tokens',
+ },
{
id: 'claude-opus-4-5',
runtimeAdapter: 'claude',