Inkling

Inkling is a large language model from Thinking Machines Lab and the model that gives the Inkling family its name. It sits at the flagship tier within that family, meaning it is positioned for the broadest and most demanding workloads Thinking Machines Lab targets rather than for cost-optimized, high-volume serving. Inkling is released as an open-weight model, so it can be hosted by multiple inference providers as well as run on self-managed infrastructure, which is why pricing for it can differ substantially from one provider to another. Inkling handles a 1,048,576-token context window (roughly 1M tokens) and accepts three input modalities: text, image, and audio. It also supports tool calling, which allows it to be wired into function-based workflows, retrieval systems, and agent frameworks. On throughput measurements collected by Artificial Analysis, Inkling generates output at about 80 tokens per second with a time to first token of roughly 2,086 ms — a profile that favors long, sustained generations over snappy interactive turnaround. In practice, the combination of a million-token context and multimodal input makes Inkling a candidate for workloads where the whole input has to be resident at once: large document sets, long transcripts paired with audio, or mixed text-and-image corpora. Compared with closed flagship models, the main structural difference is that Inkling's open weights allow self-hosting and provider choice; compared with smaller open models, the trade-off is the higher compute footprint that a flagship-tier model implies.