Skip to content

Use Claude Code, Codex, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)

Python

Preserve DeepSeek cache hits in Responses usage (#1593)

Ali Khokhard080c26f9204
diff --git a/pyproject.toml b/pyproject.toml
index 5aa8407f..f9e64c53 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "free-claude-code"
-version = "5.15.3"
+version = "5.15.4"
description = "Local proxy connecting coding agents to OpenAI-compatible AI providers"
readme = "README.md"
requires-python = ">=3.14.0"
diff --git a/src/free_claude_code/providers/deepseek/client.py b/src/free_claude_code/providers/deepseek/client.py
index d0f9a51f..df970746 100644
--- a/src/free_claude_code/providers/deepseek/client.py
+++ b/src/free_claude_code/providers/deepseek/client.py
@@ -25,6 +25,27 @@ _PROFILE = OpenAIChatProfile(
)
+def _deepseek_cache_partition(
+ usage_info: object,
+) -> tuple[int, int, int | None] | None:
+ cache_hit_tokens = usage_int(usage_info, "prompt_cache_hit_tokens")
+ cache_miss_tokens = usage_int(usage_info, "prompt_cache_miss_tokens")
+ if (
+ cache_hit_tokens is None
+ or cache_hit_tokens < 0
+ or cache_miss_tokens is None
+ or cache_miss_tokens < 0
+ ):
+ return None
+
+ prompt_tokens = usage_int(usage_info, "prompt_tokens")
+ if prompt_tokens is None or prompt_tokens < 0:
+ return cache_hit_tokens, cache_miss_tokens, None
+ if prompt_tokens != cache_hit_tokens + cache_miss_tokens:
+ return None
+ return cache_hit_tokens, cache_miss_tokens, prompt_tokens
+
+
class DeepSeekProvider(OpenAIChatProvider):
"""DeepSeek using ``https://api.deepseek.com`` Chat Completions."""
@@ -58,25 +79,18 @@ class DeepSeekProvider(OpenAIChatProvider):
finalize_deepseek_chat_body(body, reasoning)
return body
- def _anthropic_usage_fields(self, usage_info: Any) -> dict[str, int]:
- cache_hit_tokens = usage_int(usage_info, "prompt_cache_hit_tokens")
- cache_miss_tokens = usage_int(usage_info, "prompt_cache_miss_tokens")
- if (
- cache_hit_tokens is None
- or cache_hit_tokens < 0
- or cache_miss_tokens is None
- or cache_miss_tokens < 0
- ):
- return {}
+ def _cached_input_tokens(self, usage_info: object) -> int | None:
+ cache_partition = _deepseek_cache_partition(usage_info)
+ if cache_partition is None:
+ return None
+ cache_hit_tokens, _, prompt_tokens = cache_partition
+ return cache_hit_tokens if prompt_tokens is not None else None
- prompt_tokens = usage_int(usage_info, "prompt_tokens")
- if (
- prompt_tokens is not None
- and prompt_tokens >= 0
- and prompt_tokens != cache_hit_tokens + cache_miss_tokens
- ):
+ def _anthropic_usage_fields(self, usage_info: Any) -> dict[str, int]:
+ cache_partition = _deepseek_cache_partition(usage_info)
+ if cache_partition is None:
return {}
-
+ cache_hit_tokens, cache_miss_tokens, _ = cache_partition
return {
"input_tokens": cache_miss_tokens,
"cache_read_input_tokens": cache_hit_tokens,
diff --git a/src/free_claude_code/providers/openai_chat/provider.py b/src/free_claude_code/providers/openai_chat/provider.py
index 38a5afd4..f0ffd65d 100644
--- a/src/free_claude_code/providers/openai_chat/provider.py
+++ b/src/free_claude_code/providers/openai_chat/provider.py
@@ -718,14 +718,18 @@ class OpenAIChatProvider(BaseProvider):
"""Return provider-specific per-tool argument aliases for this request."""
return {}
- def _anthropic_usage_fields(self, usage_info: Any) -> dict[str, int]:
- """Split standard cached prompt tokens for final Anthropic usage."""
- prompt_tokens = usage_int(usage_info, "prompt_tokens")
- cached_tokens = nested_usage_int(
+ def _cached_input_tokens(self, usage_info: object) -> int | None:
+ """Return the provider's cached-input count from final Chat usage."""
+ return nested_usage_int(
usage_info,
"prompt_tokens_details",
"cached_tokens",
)
+
+ def _anthropic_usage_fields(self, usage_info: Any) -> dict[str, int]:
+ """Split standard cached prompt tokens for final Anthropic usage."""
+ prompt_tokens = usage_int(usage_info, "prompt_tokens")
+ cached_tokens = self._cached_input_tokens(usage_info)
if (
prompt_tokens is None
or prompt_tokens < 0
@@ -1098,14 +1102,8 @@ class _OpenAIChatStreamRunner:
usage = ChatStreamUsage(
input_tokens=completion.input_tokens,
output_tokens=completion.output_tokens,
- cached_tokens=(
- nested_usage_int(
- assembler.usage_info,
- "prompt_tokens_details",
- "cached_tokens",
- )
- or 0
- ),
+ cached_tokens=self._provider._cached_input_tokens(assembler.usage_info)
+ or 0,
reasoning_tokens=(
nested_usage_int(
assembler.usage_info,
diff --git a/tests/providers/test_deepseek.py b/tests/providers/test_deepseek.py
index 00d40527..b1b82d23 100644
--- a/tests/providers/test_deepseek.py
+++ b/tests/providers/test_deepseek.py
@@ -134,7 +134,7 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
@pytest.mark.parametrize(
- ("usage", "expected"),
+ ("usage", "anthropic_expected", "responses_cached"),
[
(
{
@@ -143,6 +143,7 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 20,
},
{"input_tokens": 20, "cache_read_input_tokens": 10},
+ 10,
),
(
{
@@ -151,6 +152,7 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 30,
},
{"input_tokens": 30, "cache_read_input_tokens": 0},
+ 0,
),
(
{
@@ -159,18 +161,22 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 0,
},
{"input_tokens": 0, "cache_read_input_tokens": 30},
+ 30,
),
(
{"prompt_cache_hit_tokens": 10, "prompt_cache_miss_tokens": 20},
{"input_tokens": 20, "cache_read_input_tokens": 10},
+ None,
),
(
{"prompt_tokens": 30, "prompt_cache_miss_tokens": 20},
{},
+ None,
),
(
{"prompt_tokens": 30, "prompt_cache_hit_tokens": 10},
{},
+ None,
),
(
{
@@ -179,6 +185,7 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 20,
},
{},
+ None,
),
(
{
@@ -187,6 +194,7 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 20.0,
},
{},
+ None,
),
(
{
@@ -195,6 +203,7 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 31,
},
{},
+ None,
),
(
{
@@ -203,6 +212,16 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": -1,
},
{},
+ None,
+ ),
+ (
+ {
+ "prompt_tokens": -1,
+ "prompt_cache_hit_tokens": 10,
+ "prompt_cache_miss_tokens": 20,
+ },
+ {"input_tokens": 20, "cache_read_input_tokens": 10},
+ None,
),
(
{
@@ -211,13 +230,15 @@ def test_responses_tool_history_keeps_deepseek_replayable_reasoning(
"prompt_cache_miss_tokens": 19,
},
{},
+ None,
),
],
)
def test_maps_only_complete_consistent_cache_usage(
- deepseek_provider, usage, expected
+ deepseek_provider, usage, anthropic_expected, responses_cached
) -> None:
- assert deepseek_provider._anthropic_usage_fields(usage) == expected
+ assert deepseek_provider._anthropic_usage_fields(usage) == anthropic_expected
+ assert deepseek_provider._cached_input_tokens(usage) == responses_cached
def test_build_request_body_openai_chat_shape(deepseek_provider):
@@ -1180,6 +1201,69 @@ async def test_stream_uses_chat_completions_and_maps_cache_usage(deepseek_provid
}
+async def test_responses_stream_maps_deepseek_cache_usage(deepseek_provider):
+ request = OpenAIResponsesRequest(model="m", input="hi")
+
+ async def fake_stream():
+ yield SimpleNamespace(
+ choices=[
+ SimpleNamespace(
+ delta=SimpleNamespace(
+ content="hello", reasoning_content=None, tool_calls=None
+ ),
+ finish_reason=None,
+ )
+ ],
+ usage=None,
+ )
+ yield SimpleNamespace(
+ choices=[
+ SimpleNamespace(
+ delta=SimpleNamespace(
+ content=None, reasoning_content=None, tool_calls=None
+ ),
+ finish_reason="stop",
+ )
+ ],
+ usage=None,
+ )
+ yield SimpleNamespace(
+ choices=[],
+ usage=SimpleNamespace(
+ completion_tokens=3,
+ prompt_tokens=30,
+ model_extra={
+ "prompt_cache_hit_tokens": 10,
+ "prompt_cache_miss_tokens": 20,
+ },
+ ),
+ )
+
+ create = AsyncMock(return_value=fake_stream())
+ with patch.object(deepseek_provider._client.chat.completions, "create", create):
+ chunks = [
+ chunk
+ async for chunk in deepseek_provider.stream_responses(
+ request, input_tokens=7, request_id="r1"
+ )
+ ]
+
+ parsed = parse_sse_text("".join(chunks))
+ completed = next(
+ event.data["response"]
+ for event in parsed
+ if event.event == "response.completed"
+ )
+ assert completed["usage"] == {
+ "input_tokens": 30,
+ "input_tokens_details": {"cached_tokens": 10},
+ "output_tokens": 3,
+ "output_tokens_details": {"reasoning_tokens": 0},
+ "total_tokens": 33,
+ }
+
+
def test_preserves_extra_body_for_openai_chat_request(deepseek_provider):
raw = {
"model": "m",
diff --git a/uv.lock b/uv.lock
index d040c576..6f169da3 100644
--- a/uv.lock
+++ b/uv.lock
@@ -607,7 +607,7 @@ wheels = [
[[package]]
name = "free-claude-code"
-version = "5.15.3"
+version = "5.15.4"
source = { editable = "." }
dependencies = [
{ name = "aiohttp" },