DeepSeek appears to be quietly testing V4.1 Pro, and early benchmark results look strong
Chinese AI lab DeepSeek seems to be giving some users a new V4.1 Pro model, and one developer's private test suggests a real jump in skill.
The short version
- DeepSeek had a short service outage, its second in a week, as the testing appeared.
- Driver Home says V4.1 Pro is in gray testing, a limited rollout to some users.
- Developer @MaxForAI compared old and new V4 Pro on his own LobeBench test set.
- The new version ranked first among Chinese models on that test and solved more Hard tasks.
- No official release date has been given, and DeepSeek has not confirmed the update.
Why it matters
If the early results hold, one of China's best-known AI labs has improved its flagship model without a big launch. For now this is a single outside test, so wait for wider release and independent checks before trusting it.
A quiet test shows up
DeepSeek, a Chinese AI lab, had another short service outage, its second in a week, as Driver Home reported. Heavy demand may explain part of it. The site argues something else is going on too: a new model called V4.1 Pro has entered gray testing.
Gray testing means a new version goes to a small slice of users before everyone gets it. The company often says nothing while it checks for problems.
What the benchmark showed
Early posts about the model used playful tasks, like drawing a cormorant riding a bike, which Driver Home dismissed as unconvincing. The more credible test came from @MaxForAI, who writes about AI and helps build LobeHub, an app for working with AI agents (assistants that take actions).
He ran the current V4 Pro and an earlier V4 Pro release through LobeBench, his own set of test questions. The current one scored clearly higher, ranked first among Chinese models on his list, and solved more of the hardest tasks.
Why consistency counts
The key detail is that the test did not reward lucky answers. Counting a task as solved if it worked at least once showed only a small gap between versions. The gap opened up when a model had to succeed three times in a row.
That suggests the model is more reliable, not just lucky. Driver Home concludes that the current V4 Pro was quietly updated and is probably the V4.1 Pro that fans were waiting for.
What to keep in mind
This is one person's private benchmark, reported by one outlet. DeepSeek has not confirmed the update or announced a date, though Driver Home says optimistic talk points to next week.
What to watch
- Will DeepSeek officially announce V4.1 Pro and give a release date?
- Will independent testers reproduce the improvement seen on LobeBench?
Try it
If you use DeepSeek, nothing needs to change today. When the release is official, test it on two or three of your own real tasks and compare with your current tool.
Sources and links
- ReportingDriver Home: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by Driver Home · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.