mirror of
https://github.com/deepseek-ai/awesome-deepseek-agent.git
synced 2026-11-05 01:07:37 +00:00
[GH-ISSUE #284] Claude Code with Deepseek Pro V4, classifier gets time-out or failed constantly in auto mode #100
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/awesome-deepseek-agent#100
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @yuhenghuang on GitHub (Jul 8, 2026).
Original GitHub issue: https://github.com/deepseek-ai/awesome-deepseek-agent/issues/284
As the title suggests, when using claude code (CLI) with deepseek pro v4 the classifier for auto mode constantly gets time-out or failed.
Even when the classifier responded successfully, the response time is significantly longer than claude code with its own models.
When using Claude Code with its own models in the same network, the auto mode classifier has close to negligible response time and much lower frequency of fail.
This issue is causing the auto mode unusable for my set up, Claude Code with Deepseek Pro V4.
I'm wondering if anyone is suffering from the same problem, or what info can I provide to help identify or address the cause.
The api.deepseek.com is resolved to
d3bbv8sr76az5s.cloudfront.netand3.173.21.63from my computer.Update:
no different from other requests on the server sideWhen thinking is enabled and the effort is high (high mode or above in my experience), it simply takes more time for the server to answer the classifier by default setting, causing the time-out errorRegarding recent release of V4 flash 0731
The auto classifier outputs false negative frequently if thinking is disabled for V4 flash. Those false negatives were not experienced with V4 Pro preview
Use auto mode + V4 flash 0731 + nonthinking classifier with caution.
Enabling thinking for V4 flash 0731 classifier would result in same timeout error as V4 pro preview
So... Although V4 flash 0731 is significantly more powerful than V4 Pro preview when thinking is enabled, without thinking it remains less competent and capable
(New!) I found that V4 flash 0731 supports low thinking level. Thus, the optimal configuration at present is auto mode + V4 flash 0731 + low thinking level classifier
Regarding recent release of V4 Pro 0813
@yesypbg-creator commented on GitHub (Jul 8, 2026):
我也遇到这个问题了，不知道怎么解决……
@yuhenghuang commented on GitHub (Jul 9, 2026):
昨天测试了一天，在相同的网络环境下：
排除了大多数变量，剩下的最可能就是server side的auto mode classifier有响应速度慢，无响应的问题了。
@WangLaoan commented on GitHub (Jul 9, 2026):
1+
@soupice commented on GitHub (Jul 9, 2026):
+1
@YanghaoZYH commented on GitHub (Jul 9, 2026):
+1， 但是cli貌似没问题
@wangyuezhen commented on GitHub (Jul 10, 2026):
+1
@anbo-hu commented on GitHub (Jul 11, 2026):
+1
@dashxio commented on GitHub (Jul 11, 2026):
测试了一个可行的解决方案，https://github.com/dashxio/deepseek-claude-proxy
@dgbaenar commented on GitHub (Jul 23, 2026):
@bcwxoandy I keep having the same problem today
@v2codes commented on GitHub (Aug 4, 2026):
+1
@AstrayWu commented on GitHub (Aug 4, 2026):
+1
@zhihuiyuze commented on GitHub (Aug 4, 2026):
https://github.com/dashxio/deepseek-claude-proxy — using this solves the problem, but it is unstable.
Can the safety classifier timeout be configured in Claude Code's automatic mode? Perhaps increasing the timeout would resolve the issue?
@yuhenghuang commented on GitHub (Aug 5, 2026):
As far as I researched recently, there's no way to change the timeout of classifier in claude code, which is observed to be around 30 seconds. The distribution of classifier's response time with thinking enabled is around 25 ~ 35 seconds during peak hours. Lifting the timeout of the classifier would definitely solve the problem, but the reality is that we can't for the time being.