Thinking Machines Inkling
Thinking Machines' open-weight multimodal mixture-of-experts model (975B total, 41B active parameters) for reasoning, coding and agent work, reading text, images and audio.
Thinking Machines's Inkling is a cloud text model. It is multimodal: alongside a text prompt it accepts images and audio, then replies with generated text. Its context window handles up to 524K input tokens and up to 472K output tokens, and it supports adjustable reasoning effort for harder problems. Generation can be tuned with system instructions, temperature, and top-p sampling. It runs through OpenRouter using your own API key, from $1.00 per million input tokens.
- Released
- Jul 2026
- Pricing
- $1.00 / 1M in · $4.05 / 1M out
- Inputs
- Text, Images, Audio
- Context window
- 524K in · 472K out
- Reasoning
- Adjustable effort
- Controls
- System prompt, Temperature, Top-p
Samples
Generated with Inkling via OpenRouter. The same three prompts run against every text model in the catalog, shown verbatim, so the only thing that changes between two models’ answers is the model.
Explain what an API is to someone who has never written code. Do not use a restaurant or waiter analogy. Plain prose only — no headings, bullet points, or markdown formatting. Maximum 50 words.
An API is a messenger that lets different computer programs talk to each other. One program asks for specific information or actions through clear rules, and the other responds automatically. It hides all the complex inner workings so they can cooperate smoothly without understanding each other's internal code.
A two-person startup is deciding whether to build their own user authentication or use a hosted provider. Give them a clear recommendation, then the single strongest argument against your own recommendation. Plain prose only — no headings, bullet points, or markdown formatting. Maximum 150 words.
Use a hosted authentication provider. For a two-person team, building secure login systems is a dangerous distraction; you must handle password hashing, session management, multi-factor authentication, breach recovery, and constant security patches, all of which require specialized expertise you likely do not have. A hosted service lets you ship faster, reduces liability, and handles compliance out of the box. The single strongest argument against this recommendation is vendor lock-in combined with scaling costs: as your user base grows, per-user pricing can become a significant recurring expense, and migrating away from a proprietary system later requires painful data migration and potential downtime, especially if you eventually need custom authentication flows that the provider does not support. Still, for an early-stage startup, the speed and reduced risk overwhelmingly outweigh these future concerns until you have dedicated security engineers and revenue to justify owning the stack.
Write a briefing for a non-technical executive explaining why their company's spending on AI is hard to predict, and what they can do about it. Cover at least three distinct causes. Plain prose only — no headings, bullet points, or markdown formatting. Write exactly 205 words.
Artificial intelligence budgets are slippery because the technology does not behave like traditional software with fixed licensing fees. First, usage costs scale with real-world interaction rather than seat count. Every customer query or document summary draws on cloud computing priced by the token, so a viral marketing campaign or a suddenly popular feature can multiply the bill overnight without anyone ordering more licenses. Second, the field moves faster than procurement cycles can track. A model that seemed state-of-the-art in January may be obsolete by June, yet contracts often lock you into specific vendors, forcing expensive migrations or shadow spending on newer tools that teams adopt informally to stay competitive. Third, hidden labor costs accumulate around integrating messy internal data and monitoring outputs for errors or bias, work that rarely appears in vendor quotes but consumes engineering and legal time continuously. To regain control, shift from fixed capital budgets to flexible operational allocations tied to actual consumption metrics, build vendor-agnostic architectures so you can switch providers without rebuilding systems, and create a cross-functional review board that approves experiments only after estimating both direct compute costs and internal labor required to deploy them safely. Treat AI spending as an ongoing operational risk rather than a one-time purchase.
Thinking Machines
Thinking Machines Lab builds open-weight multimodal models, including Inkling and Inkling Small, which read text, images and audio, and Tinker, a fine-tuning API.
thinkingmachines.ai ↗