Inkling 2 Small vs MiMo-V2.6 Distill Qwen 9B

Inkling 2 Small

• +6.7 on 12 November 2026 (D to C): product

MiMo-V2.6 Distill Qwen 9B

• no change on 3 November 2026: product

Inkling 2 Small

Thinking Machines' newest open model, a 594B mixture-of-experts generalist with images in, several tool calls a turn and per-request thinking effort. The NVFP4 checkpoint needs about 472 GB, and no released runtime runs it yet, so today that means vLLM's or SGLang's own images, or Tinker and Together hosted. The use policy binds on download and can change.

MiMo-V2.6 Distill Qwen 9B

A 9B cousin tuned on MiMo-V2.6's agent work for coding, terminal tasks and visual coding, released as a base for RL research. Ollama's library build needs about 8 GB at 4-bit, while Xiaomi ships only BF16. The main risks are no safety results while a seventh of its tuning is offensive security, and nobody at Xiaomi answering reports.

Inkling 2 Small

1. No released runtime runs it yet. Serve the NVFP4 checkpoint from SGLang's dev-inkling-2-small image with the cookbook's recipe for your hardware, or from vLLM's vllm/vllm-openai:inkling-2-small image
2. In vLLM pass --reasoning-parser inkling --enable-auto-tool-choice --tool-call-parser inkling. In SGLang the tool-call parser is inkling2 and the reasoning parser inkling
3. Transformers' main branch loads only thinkingmachines/Inkling-2-Small-MXFP8, the 622.8 GB checkpoint, and no release has the code yet
4. Pass reasoning_effort as none, minimal, low, medium, high (the default), xhigh or max, or a number from 0 to 0.99
5. Effort none conditions the model toward no reasoning but doesn't forbid it, and higher effort needs a larger max_tokens
6. Tool call arguments must reach the template as an object, not a JSON string, or it raises an error
7. Images should be 40 to 4096 px a side. The model takes no audio, and the template raises an error for an audio part
8. Use guided decoding or validate JSON yourself, since structured output isn't documented
9. Without the hardware, call it on Tinker's OpenAI-compatible endpoint with a Tinker key, at $0.30 input and $1.00 output per million tokens as read on 6 November 2026
10. Read the Model Acceptable Use Policy before deploying, it binds on download and bars some uses that are legal

MiMo-V2.6 Distill Qwen 9B

1. Quickest start is ollama run mimo-v2.6:9b, Ollama's 6.6 GB q4_K_M build with text and image input
2. Serve it as the card does, sglang serve --model-path XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B --reasoning-parser mimo, and add --tool-call-parser mimo for parsed tool calls
3. SGLang's mimo reasoning parser expects thinking to be asked for, so pass chat_template_kwargs with enable_thinking true, as the card does, and read reasoning_content
4. Calls come back as function and parameter blocks inside tool_call, the same shape vLLM's and SGLang's qwen3_coder parsers read
5. The generation config samples at temperature 0.6, top_k 20 and top_p 0.95
6. Images and video go in through Qwen3.5's processor; the template's audio tokens have no audio encoder behind them
7. The MTP head isn't in the files, so skip MTP speculative decoding even though config.json lists a layer
8. It's an SFT checkpoint meant for further RL, and discussions report runaway generation on llama.cpp, so cap max_tokens
9. If you redistribute the weights, include the Apache 2.0 text of Qwen3.5-9B, which section 4(a) asks for and the repository lacks