Inkling 2 Small vs Kolibri-1

Inkling 2 Small

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

Kolibri-1

• no change on 3 November 2026: product
• -14.8 on 21 October 2026 (BB to C): checklist

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.

Kolibri-1

Kolibri-1 is a German and English reasoning model with a documented tool-call format, switchable thinking and a 262,144-token context, under Apache 2.0. Aleph Alpha's FP8 build needs about 95 GB and runs on vLLM and Transformers 5.20; llama.cpp and Ollama 0.40.14 need a GGUF. The main risk is no published safety results or security contact, and few repository answers.

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

Kolibri-1

1. Start it with pip install 'aleph-alpha-inference>=1' then vllm serve Aleph-Alpha/Kolibri-1 --kv-cache-dtype fp8 --reasoning-parser kolibri1 --tool-call-parser kolibri1 --enable-auto-tool-choice
2. Without the kolibri1 tool-call parser and enable-auto-tool-choice, tool calls come back as text in tool_call tags
3. The plugin supports one vLLM minor version, 0.29. vLLM 0.33.0 loads the model without it, but has no kolibri1 tool or reasoning parser
4. Thinking is on by default at high effort. Pass reasoning_effort none or enable_thinking false in chat_template_kwargs for short tasks
5. Recommended sampling is temperature 1.0, top_p 0.97 and top_k 128
6. Past 262,144 tokens, add --max-model-len 1048576 and the max_position_embeddings override in --hf-overrides, though Aleph Alpha recommends staying at or below 262,144
7. With Transformers 5.20 or later, tokenizer.parse_response returns the answer in content and the reasoning in thinking
8. On the reference machine the FP8 build fits by size, but no document confirms vLLM runs FP8 on gfx1151, and Aleph Alpha ships no GGUF for llama.cpp or Ollama
9. Validate tool results in your own code, as the card asks