Thinking Machines: Inkling

Thinking Machines: Inkling is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab. It has 41B active parameters out of 975B total, and it is designed for general-purpose reasoning, coding, agentic and tool-use systems. On api14 it is listed under the model id thinkingmachines/inkling and falls into the text category.

The model accepts text, image and audio as input and produces text as output, so it can work across mixed-media prompts rather than text alone. Its context window is 524,288 tokens, and the maximum output is 471,859 tokens. That combination of a long input window and a large output budget makes it suitable for tasks where a lot of material has to be read and a substantial amount of text has to be written back in a single request.

Practical uses follow from those capabilities. Long documents, large codebases and extended transcripts can be passed in without heavy chunking, which helps with code generation and review, multi-step reasoning over lengthy source material, and agentic or tool-use systems that need to keep a lot of state in context. Because images and audio can be supplied alongside text, the same model can handle prompts that mix a screenshot or a recording with written instructions, then return a text answer. The reasoning and tool parameters also point to workflows where the model plans, calls tools and reports results.

You call it through api14's OpenAI-compatible API using the model id thinkingmachines/inkling. The supported request parameters are frequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, seed, stop, temperature, tool_choice, tools, top_k and top_p. Tool use is configured with tools and tool_choice, output length with max_tokens, sampling behaviour with temperature, top_p, top_k, min_p and the penalty parameters, and reasoning behaviour with reasoning, reasoning_effort and include_reasoning. A seed can be set for reproducible sampling, and stop sequences can be supplied to end generation at chosen points.

The main constraint to keep in mind is that output is text only, so tasks that require generating images or audio are outside its scope. If your workload is purely text and does not need the multimodal inputs or the long context, a smaller text-only model may be a simpler fit. For mixed-media prompts, very long inputs or tool-driven agents, Inkling is the appropriate choice among the models available here.