Kat Coder Pro V2.5
Kat Coder Pro V2.5 is the flagship of KwaiPilot's KAT-Coder V2.5 generation. It scores 65.2 on SWE-Bench Pro and 94.9 on PinchBench, with a context window of 256K tokens and output up to 80K tokens. Your use is subject to KwaiPilot's Terms & Privacy Policies.
import { streamText } from 'ai'
const result = streamText({ model: 'kwaipilot/kat-coder-pro-v2.5', prompt: 'Why is the sky blue?'})Playground
Try out Kat Coder Pro V2.5 by KwaiPilot. Usage is billed to your team at API rates. Free users (those who haven't made a payment) get $5 of credits every 30 days.
Kat Coder Pro V2.5
Kat Coder Pro V2.5
Providers
Route requests across multiple providers. Copy a provider slug to set your preference. Visit the docs for more info. Using a provider means you agree to their terms, listed under Legal.
P50 throughput on live AI Gateway traffic, in tokens per second (TPS). Visit the docs for more info.
P50 time to first token (TTFT) on live AI Gateway traffic, in milliseconds. View the docs for more info.
Direct request success rate on AI Gateway and per-provider. Visit the docs for more info.
More models by KwaiPilot
| Model |
|---|
| 256K | 0.8s | 344tps | $0.15/M | $0.60/M | Read:$0.03/M Write:— | — | +1 |  | 07/10/2026 | ||||
| 256K | 0.9s | 114tps | $0.30/M | $1.20/M | Read:$0.06/M Write:— | — | | 03/27/2026 | |||||
| 256K | 1.8s | 107tps | $0.30/M | $1.20/M | Read:$0.06/M Write:— | — |  | 11/09/2025 |
About Kat Coder Pro V2.5
Kat Coder Pro V2.5 is the flagship model in KwaiPilot's KAT-Coder V2.5 generation, released July 10, 2026. A task starts from a natural-language issue and ends with a verified change. Kat Coder Pro V2.5 locates the relevant code across files, follows the repository's existing conventions, edits, and runs tests to confirm the change holds.
On SWE-Bench Pro, which measures repository-level software engineering, Kat Coder Pro V2.5 scores 65.2 and ranks second in KwaiPilot's published comparison, behind Opus 4.8 at 69.2. On PinchBench, an agentic tool-use benchmark, Kat Coder Pro V2.5 records 94.9, the top result in that comparison. It also scores 53.1 on KAT Code Bench, 60.7 on Terminal-Bench 2.1, and 50.3 on SciCode. Every model in that comparison ran under the same Claude Code harness, tool set, context budget, and decoding configuration.
Kat Coder Pro V2.5 supports streaming output, context caching, function calling, and MCP (Model Context Protocol), and KwaiPilot lists coverage of more than 20 mainstream programming languages including C, C++, Java, and Python. Cached input reads bill at $0.15 per million tokens, which matters for agent loops that resend the same repository context turn after turn. Route requests through AI Gateway to reach Kat Coder Pro V2.5 via StreamLake. Product details sit at .
What To Consider When Choosing a Provider
- Configuration: Kat Coder Pro V2.5 is the higher-capability half of the KAT-Coder V2.5 pair.
kat-coder-air-v2.5shares the same context window, output ceiling, and tool surface, and KwaiPilot publishes a lower SWE-bench figure for it. Compare the two against your own task mix before you commit. - Configuration: Agent sessions accumulate context quickly. The window of 256K tokens covers large multi-file work, but caching repeated context is what keeps per-turn cost predictable once a session runs long. Check the pricing panel on this page for current input, output, and cached read rates.
- Configuration: Benchmark numbers here come from KwaiPilot's own evaluation under a fixed harness. Treat them as a starting point and measure resolution rate on your repositories before you route production traffic.
- Zero Data Retention: AI Gateway does not currently support Zero Data Retention for this model. See the documentation for models that support ZDR.
- Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.
When to Use Kat Coder Pro V2.5
Best for
- Autonomous Issue Resolution: Locate, edit, and verify changes inside a real repository
- Long-Horizon Engineering: Tasks spanning many files and dozens of tool calls per session
- MCP Agent Harnesses: Function calling and MCP support fit existing agent tooling
- Large Repository Sessions: A window of 256K tokens holds multi-file diffs and session history
- Cached Agent Loops: Context caching keeps resent repository context affordable
Consider alternatives when
- Air Tier Suffices:
kat-coder-air-v2.5shares the context window and output ceiling at lower cost - Multimodal Input Needed: Screenshots and diagrams require a model that accepts image input
- Terminal and Scientific Coding: Scores of 60.7 on Terminal-Bench 2.1 and 50.3 on SciCode trail the coding results
- General-Purpose Work: Writing and open-ended analysis sit outside a coding model's scope
Conclusion
Kat Coder Pro V2.5 is KwaiPilot's strongest KAT-Coder release for repository-level work, with 65.2 on SWE-Bench Pro and the top PinchBench result in its published comparison. Pair it with kat-coder-air-v2.5 on routine tasks and route both through AI Gateway so switching tiers is a model identifier change.