Model Terminal

KAT-Coder-Pro V2.5

KAT-Coder-Pro V2.5 is an agentic coding large language model developed by Kwaipilot (Kuaishou's AI research division), designed for autonomous software engineering tasks including bug fixing, repository-level code changes, and multi-step tool-use workflows. It features a 256K context window and is benchmarked specifically on agentic coding evaluations such as SWE-Bench Pro. Source: supabase.

Value score
Not scored
Context
tokens
Max output
tokens
Price

Capability radar

Identity

Developer
Kwaipilot
Openness
Modalities
Release
Knowledge cutoff
Deprecation
API docs

Benchmarks

No benchmark scores yet.

Compare nearby

Pricing

Input / 1M
Output / 1M
Speed