X
#KAT-Coder-Pro V1
green igloo
X
rustic shale
their last model wasnt amazing
errant parcel
The speed of api was a little bit slow. Anyway, performance and intelligence is great, respect
dusk cave
What is the size of the KAT-Coder-Pro model? How does it compare to other, tooling benchmarks ?
honestly, the performance feels decent enough.
unborn plume
https://x.com/KwaiAICoder/status/1981346919680213306
already added
here in the list
weary spade
this model is pretty good, it made this which i think is pretty good
graceful echo
their last model wasnt amazing
f you have more suggestions, don't hold back! We're all ears and ready to polish things up based on your feedback.
graceful echo
The speed of api was a little bit slow. Anyway, performance and intelligence is ...
We've noted your feedback and look forward to seeing you build more projects with KAT-Coder-Pro V1!
graceful echo
this model is pretty good, it made this which i think is pretty good
Appreciate the share! Let's create something amazing together.
coral heart
KAT-Coder-Pro V1：1201 HUGE UPDATE!
The latest version of KAT-Coder-Pro V1 （1208）scores 64 on the Artificial Analysis Intelligence Index, ranking 9th globally among all models, and the 1st position among non-reasoning models!
Now, here are the primary improvements of KAT-Coder-Pro V1：
🚀 Exceptional Agentic Interaction
Deeply optimized integration with dozens of mainstream Agent tools such as Claude Code, Kilo Code, Roo Code, Cline, and Zed, significantly improving interaction smoothness and response precision across different development environments.
🧠 Enhanced Code Reasoning & Tool Invocation
Further improved Agentic Coding capabilities, strengthening the model's ability to utilize Coding Tools and Web Search Tools to solve real-world problems in various scenarios.
🎨 Superior Frontend Code Generation
Significantly enhanced frontend aesthetic appeal and code accuracy for HTML, CSS, and JavaScript by leveraging a new generative reward model.
🛡️ Improved Tool Stability
Reduced the overall error rate of tool (API) calls, boosting the stability of complex task workflows.
📈 General Task Capability Boost
Achieved a significant performance improvement, pushing the AA comprehensive score past 64.
No change to API usage:
● StreamLake：https://www.streamlake.ai/product/kat-coder#packages
Openrouter: https://openrouter.ai/kwaipilot/kat-coder-pro:free
rustic shale
definitely an improvement in my harness, no longer hallucinating grep tools and so on
does seem slightly less agentic than some other models in the sense you need to tell it to do the task
model quickly degrades in tool calling past like 128k~ though, starts hallucinating file paths and arguments
rustic shale
i asked to update my agents file and it reached about 210k ctx then the requests stopped working (my cli sais 200k cause im too lazy to update it but the model has 256k)
sleek raft
X
glossy inlet
So openrouter endpoint auto updated?
glossy inlet
model sucks, benchmaxxed. worse than glm air
graceful echo
So openrouter endpoint auto updated?
Yes, it can update automatically.
graceful echo
model sucks, benchmaxxed. worse than glm air
Feel free to share your assessment results — we’re always glad to listen and optimize.
graceful echo
i asked to update my agents file and it reached about 210k ctx then the requests...
We have received your issue and are working on identifying and optimizing it. Thank you very much for your feedback.
glossy inlet
Feel free to share your assessment results — we’re always glad to listen and opt...
sure if i get time. but keep cooking think it has strong potential