Inkling-Small vs Qwen3.6-27B

Inkling-Small

A mixture-of-experts generalist with images and audio in, several tool calls a turn and per-request thinking effort. It needs about 205 GB for Thinking Machines' NVFP4 build on recent NVIDIA or AMD datacentre GPUs, and Tinker or DeepInfra host it. The main risks are a use policy that binds on download and can change, and no outside safety results.

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.

Inkling-Small

1. Start the NVFP4 build with SGLang's recipe, sglang serve --trust-remote-code --model-path thinkingmachines/Inkling-Small-NVFP4 --tp 8 --enable-multimodal --reasoning-parser inkling --tool-call-parser inkling, on Blackwell for W4A4
2. Use the inkling tool and reasoning parsers in vLLM or SGLang, or Thinking Machines' tml-renderers package to render prompts and parse calls yourself
3. Pass reasoning_effort as none, minimal, low, medium, high (the default) or max, or a number from 0 to 0.99. The repository's template has no xhigh preset and raises an error for it, though Thinking Machines' docs and SGLang's cookbook list xhigh
4. Effort none conditions the model toward no reasoning but doesn't forbid it, and higher effort needs a larger max_tokens
5. Tool call arguments must reach the template as an object, not a JSON string, or it raises an error
6. Images should be 40 to 4096 px a side and audio 16 kHz WAV under two minutes, per the card
7. Use guided decoding or validate JSON yourself, since structured output isn't documented
8. Read the Model Acceptable Use Policy before deploying, it binds on download and bars some uses that are legal

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