KAT-Coder-V2.5-Dev by Kwaipilot — a 35B-A3B sparse MoE agentic coding model achieving 69.40 SWE-bench Verified, quantized for Apple Silicon using the BaseQuant_XL 6/8-bit recipe.
KAT-Coder-V2.5-Dev is a post-trained MoE built on Qwen3.6-35B-A3B via two-stage SFT (127K examples) + RL (10 epochs) with token-level consistency (TITO), truncated importance sampling (TIS), and hierarchical reward shaping from harness execution feedback. It ships text-only weights (no vision tower) despite the Qwen3_5MoeForConditionalGeneration architecture — the MLX conversion uses mlx_vlm with strict=False to skip the 333 missing vision_tower.* parameters.
About XL Quantization
BaseQuant_XL is a fully data-agnostic, static quantization. No calibration dataset, no sensitivity analysis, no importance matrix. Precision is allocated purely by architectural role — routing-critical layers get higher precision, bulk expert parameters get lower precision. The result is a transparent, faithful capture of the source model.
Data-dependent calibration quantizations (iMatrix, AWQ, GPTQ, oQ, oQ4e, etc.) use a calibration set to guide bit allocation. This can produce a skewed representation of the model: domains well-represented in the calibration data (English, popular topics, public or leaked benchmarks) are preserved better, while underrepresented domains (non-English languages, niche use cases, your own data) are preserved worse. XL avoids this trade-off entirely — it generalizes honestly because it is never fit to any particular data distribution.
Quickstart
pip install -U mlx-lm
python -m mlx_lm.generate --model leonsarmiento/KAT-Coder-V2.5-Dev-6bit-XL-mlx --max-tokens 256 --temperature 1.0 --top-p 0.95 --prompt "Implement an LRU cache in Python with O(1) get/put."
Works with LM Studio. Thinking mode is on by default (emits <think>...</think>). Pass enable_thinking=False for direct answers.
Quantization Strategy
BaseQuant_XL recipe — precision is allocated by layer importance, not applied uniformly:
Bulk parameters — 256 experts with only 8 active per token; redundancy absorbs quantization noise. 6-bit is the sweet spot for routed experts (higher bits can cause overthinking)
Bits per weight: ~6.9 · Total size: ~28 GB (6 shards) · Group size: 64
Recommended Inference Parameters
Thinking Mode (default)
Parameter
Value
temperature
1.0
top_p
0.95
top_k
20
presence_penalty
1.5
reasoning_parser
qwen3
Instruct Mode (non-thinking)
Parameter
Value
temperature
0.7
top_p
0.8
top_k
20
presence_penalty
1.5
enable_thinking
false
preserve_thinking is supported — set chat_template_kwargs: {"preserve_thinking": true} to retain thinking traces from historical messages (beneficial for agentic scenarios).
Chat Template
The chat_template.jinja and tokenizer_config.json were updated on 2026-07-30 to match the latest chat template published by Kwaipilot in Kwaipilot/KAT-Coder-V2.5-Dev. This update enforces that system messages must appear at the beginning of the conversation — mid-conversation system messages now raise an exception instead of being silently rendered.
Model Overview
Property
Value
Architecture
Qwen3.6 MoE (35B-A3B) — text-only weights
Parameters
35B total / ~3B active per token
Experts
256 (8 routed + 1 shared)
Context window
262,144 tokens native (~1M with YaRN)
Attention
Hybrid linear + full (full every 4th layer)
Thinking
On by default (<think>...</think>)
Tool calling
XML-style via qwen3_coder parser
License
Apache-2.0
MLX Benchmarks (n=50, instruct mode)
Benchmark
KAT-Coder-V2.5-Dev 6bit-XL
Qwen3.6-35B 6bit-XL
Gemma-4 26B 6bit-XL
MMLU
70%
64%
76%
MMLU_PRO
64%
64%
82%
ARC_CHALLENGE
90%
90%
90%
HUMANEVAL
92%
78%
98%
MBPP
78%
78%
82%
Average
78.8%
74.8%
85.6%
KAT-Coder leads Qwen3.6 on 4 of 5 benchmarks — the RL training for agentic coding shows in HUMANEVAL (+14 points). MMLU_PRO trails Gemma-4 significantly (64% vs 82%); KAT-Coder is specialized for SWE, not broad knowledge. Coding is strong but Gemma-4 remains overall champion.
KAT-Coder is the reasoning efficiency champion — 9× less reasoning per correct answer than Gemma-4, and 20× less than Salience. Its SWE-focused RL training produces extremely terse, decisive reasoning chains. The one genuine failure (Q636) was an algebra error, not a reasoning deficit. KAT doesn't overthink — it computes, verifies, and answers.
Preserve Thinking
KAT-Coder is a Qwen3.6 model and supports preserve_thinking for agentic multi-turn workflows. Set chat_template_kwargs: {"preserve_thinking": true} to retain thinking traces from historical messages — essential for long-horizon SWE tasks where the model needs to recall its own reasoning from prior turns.
Verification (number-guessing game, 4 turns): With preserve_thinking enabled, KAT stays perfectly consistent — picking a secret number in turn 1 and giving correct higher/lower hints in all subsequent turns (1.4–1.6s each). The model reads its own prior <think> blocks rather than re-deriving the answer. Without preserve_thinking, the model re-picks a new secret number each turn and reconstructs consistency from visible conversation history — which works for simple games but risks drift in complex agentic tasks.
SWE-bench SOTA — 69.40 Verified at this scale class
Deep reasoning — structured <think>...</think> chains with one-token switch to disable
Tool calling — RL-trained for reliable, non-degenerate tool-use patterns
Flight Simulator — Coded by This Model
A real-world coding benchmark: each model was prompted to write a complete flight simulator from scratch. The resulting code was rendered and recorded.
flightsimulator
KAT-Coder made by Kwaipilot (KwaiKAT Team). MLX quantization by leonsarmiento.
