Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

fix(providers): update DeepSeek V3 context limit from 128K to 1M (#1038) (#1137)

Shengbo_Wangbcabc5cb11c7
diff --git a/headroom/providers/openai_compatible.py b/headroom/providers/openai_compatible.py
index daf8a3fd9..8a590d51c 100644
--- a/headroom/providers/openai_compatible.py
+++ b/headroom/providers/openai_compatible.py
@@ -84,10 +84,11 @@ _DEFAULT_CONTEXT_LIMITS: dict[str, int] = {
"qwen2-72b": 32768,
"qwen2.5": 131072,
# DeepSeek
- "deepseek": 32768,
- "deepseek-coder": 16384,
+ "deepseek": 1048576,
+ "deepseek-coder": 128000,
"deepseek-v2": 128000,
- "deepseek-v3": 128000,
+ "deepseek-v3": 1048576,
+ "deepseek-v4": 1048576,
# Yi
"yi": 32768,
"yi-34b": 32768,
diff --git a/tests/test_providers/test_universal.py b/tests/test_providers/test_universal.py
index 5a96800fb..9f34e622e 100644
--- a/tests/test_providers/test_universal.py
+++ b/tests/test_providers/test_universal.py
@@ -81,6 +81,14 @@ class TestOpenAICompatibleProvider:
limit = provider.get_context_limit("llama-3.1-8b")
assert limit == 128000
+ def test_get_context_limit_deepseek_v3_is_1m(self):
+ """DeepSeek V3/V4 support 1M context, not 128K (#1038)."""
+ provider = OpenAICompatibleProvider()
+ assert provider.get_context_limit("deepseek-v3") == 1048576
+ assert provider.get_context_limit("deepseek-v4") == 1048576
+ assert provider.get_context_limit("deepseek") == 1048576
+ assert provider.get_context_limit("deepseek-v2") == 128000
+
def test_get_context_limit_unknown_model(self):
"""Test context limit for unknown models (defaults to 128K)."""
provider = OpenAICompatibleProvider()