kat-dev

by Qwen

KAT-Dev (32B) is an open-source 32B parameter model specifically designed for software engineering tasks. It achieved a 62.4% resolution rate on the SWE-Bench Verified benchmark, ranking fifth among all open-source models of various scales. The model is optimized through multiple stages, including intermediate training, supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT), as well as large-scale agent reinforcement learning (RL). Based on Qwen3-32B, its training process lays the foundation for subsequent fine-tuning and reinforcement learning stages by enhancing fundamental abilities such as tool usage, multi-turn interaction, and instruction following. During the fine-tuning phase, the model not only learns eight carefully curated task types and programming scenarios but also innovatively introduces a reinforcement fine-tuning (RFT) stage guided by human engineer-annotated “teacher trajectories.” The final agent reinforcement learning phase addresses scalability challenges through multi-level prefix caching, entropy-based trajectory pruning, and efficient architecture.

API Pricing

Input$0.14 / 1M tokens
Output$0.55 / 1M tokens

Specifications

Context window128,000 tokens
Modalitiestext
Featurestool calling

FAQ

What is kat-dev?

KAT-Dev (32B) is an open-source 32B parameter model specifically designed for software engineering tasks. It achieved a 62.4% resolution rate on the SWE-Bench Verified benchmark, ranking fifth among all open-source models of various scales. The model is optimized through multiple stages, including intermediate training, supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT), as well as large-scale agent reinforcement learning (RL). Based on Qwen3-32B, its training process lays the foundation for subsequent fine-tuning and reinforcement learning stages by enhancing fundamental abilities such as tool usage, multi-turn interaction, and instruction following. During the fine-tuning phase, the model not only learns eight carefully curated task types and programming scenarios but also innovatively introduces a reinforcement fine-tuning (RFT) stage guided by human engineer-annotated “teacher trajectories.” The final agent reinforcement learning phase addresses scalability challenges through multi-level prefix caching, entropy-based trajectory pruning, and efficient architecture.

What is the context length of kat-dev?

kat-dev has a 128,000 token context window.

How much does kat-dev cost?

On AIHubMix, kat-dev costs $0.14 per million input tokens and $0.55 per million output tokens.

What modalities does kat-dev support?

kat-dev accepts text input.

What features does kat-dev support?

kat-dev supports tool calling. Per-protocol parameter support is listed in the capability table on this page.

How do I call kat-dev via API?

kat-dev is available through the AIHubMix unified API. The API is OpenAI-compatible: point your OpenAI SDK at https://aihubmix.com/v1, use your AIHubMix API key, and set the model name to kat-dev — no other code changes needed.

Who develops kat-dev?

kat-dev is developed by Qwen. AIHubMix aggregates it alongside models from other providers behind one API and one bill.

More from Qwen

qwen3.8-max-preview

by Qwen

Qwen 3.8 Max Preview(Qwen3.8-Max-Preview) is the latest-generation foundation model in…

$0.17/1M in · $0.51/1M out
983,616 tokens context

qwen-audio-3.0-tts-flash

by Qwen

qwen-audio-3.0-tts-flash is a high-performance speech synthesis large model optimized for…

$14.2/1M in · $14.2/1M out

qwen-audio-3.0-tts-plus

by Qwen

qwen-audio-3.0-tts-plus is a high-performance speech synthesis large model designed for…

$15/1M in · $15/1M out

happyhorse-1.1-i2v

by Qwen

HappyHorse-1.1-I2V supports image-to-video generation, further enhancing visual texture…

$2/1M in

happyhorse-1.1-r2v

by Qwen

HappyHorse-1.1-R2V supports reference-based video generation, further improving the…

$2/1M in

happyhorse-1.1-t2v

by Qwen

HappyHorse-1.1-T2V supports text-to-video generation, further enhancing text semantic…

$2/1M in

Use kat-dev via the AIHubMix unified API — one interface for every major LLM.