Kimi K3 vs Qwen3.6-27B

Kimi K3

Kimi K3 is a 2.8T multimodal agent model with a 1M-token context, parallel tool calls and per-request effort, for long coding and research sessions. Its 1,561 GB of files need a multi-node GPU cluster, so most agents will call a hosted copy. The main risks are Moonshot's Python running at load and prompt time, and no answers in its repository.

Qwen3.6-27B

A capable dense cousin for agentic coding and tool calls with images and video in, running in every major runtime. Ollama's 4-bit build needs about 22 GB, and three providers host it. The main risk is silence, with no safety results, no agent guidance, no answers in its repository and Qwen3.8-27B now ahead of it.

Kimi K3

1. Start it with SGLang on a node of 8 B300s, sglang serve --trust-remote-code --model-path moonshotai/Kimi-K3 --tp-size 8 --context-length 65536 --mem-fraction-static 0.82 --reasoning-parser kimi_k3 --tool-call-parser kimi_k3 --port 30000
2. Use the kimi_k3 tool-call and reasoning parsers in vLLM or SGLang, and validate tool-call arguments against the schema, since vLLM's recipe says K3 sometimes emits a format its parser doesn't expect
3. Pass the whole assistant message back every turn, reasoning_content and tool_calls included, or output becomes unstable
4. Thinking is always on; set reasoning_effort to low or high when max is too slow
5. Sampling per the card is temperature 1.0, with top-p 0.95 for single-step tasks and 1.0 for agentic ones
6. Don't hand an ongoing session from another model to K3, and state boundaries in the system prompt, since Moonshot warns it can act on its own when intent is unclear
7. Without the hardware, call it through Moonshot's API or an inference provider listed on its Hugging Face page

Qwen3.6-27B

1. Quickest start is ollama run qwen3.6:27b, about 18 GB of files with the vision projector
2. Alibaba's FP8 build with the card's parsers is vllm serve Qwen/Qwen3.6-27B-FP8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder, vllm 0.19.0 or later
3. Use the qwen3_coder tool parser, the template writes calls as function and parameter blocks inside tool_call
4. Thinking is on by default, pass "chat_template_kwargs": {"enable_thinking": false} for short tasks, and /think and /nothink aren't supported
5. "preserve_thinking": true in chat_template_kwargs keeps earlier reasoning in context, which helps long agent sessions but grows the prompt
6. Sampling per the card is temperature 1.0, top_p 0.95, top_k 20 when thinking, and temperature 0.7, top_p 0.8, presence_penalty 1.5 without
7. Keep at least 128K of context if you rely on thinking, and turn on YaRN only beyond 262,144 tokens
8. Validate any JSON it returns or use the runtime's guided decoding, since structured output isn't documented
9. Qwen3.8-27B (listed as qwen3-8-27b) is the newer model at this size on the same runtimes