Key Takeaways
- Liquid AI d1-ultra is a multimodal decision model released on October 15, 2026, designed for hard decision problems rather than general chat.
- It supports 44 languages and leads all eight categories of Liquid AI's own 112-evaluation suite, and it scored 65 on the Decision Index.
- Access is offered through an API; the quoted price is about $0.042 per million input tokens with output tokens not billed, but confirm current pricing on liquid.ai.
- It is best suited for developers and teams building decision-support workflows, not for casual chatbot use.
Quick Facts
| Type | Multimodal decision model |
|---|---|
| Devices | API access (cloud) |
| Price | ~$0.042 per 1M input tokens; output not billed |
| Languages | 44 languages supported |
| Highlight | Leads all 8 categories of Liquid's 112-eval suite |
| Learning curve | Moderate; API integration required |
| Category | AI & Models |
| Creator | Liquid AI |
| Official site | liquid.ai ↗ |
| Heat | 🔥 301.1K Heat |
| Last updated |
What Liquid AI d1-ultra Actually Is
Liquid AI d1-ultra is a new AI model from Liquid AI, released on October 15, 2026. Unlike most recent model launches that position themselves as general-purpose chatbots, d1-ultra is explicitly framed as a decision model built for hard decision problems. Liquid AI calls it the most powerful decision model it has released so far. The model is multimodal, meaning it can process more than just text, and it supports 44 languages. This decision-first positioning sets it apart from the crowded field of conversational AI.
The model scored 65 on the Decision Index, a benchmark that evaluates how well AI systems perform on decision-oriented tasks. It also leads all eight categories of Liquid AI's own 112-evaluation suite. That suite is internal to Liquid AI, so the category-leading claim should be read as the company's own assessment rather than an independent third-party result. The Decision Index score, however, is a public benchmark, which gives developers an external reference point. Together, these results suggest d1-ultra is optimized for structured reasoning and choice-making rather than open-ended chat.
Access to d1-ultra is offered through an API, which means it is aimed at developers and teams building applications rather than end users looking for a consumer app. There is no dedicated product page for d1-ultra yet; the official site is liquid.ai, where you can check for updates. The API-only availability is consistent with Liquid AI's focus on enterprise and developer use cases. If you are evaluating models for a decision-support tool, d1-ultra is worth a look, but you will need to integrate it yourself.
- Decision-first design
d1-ultra is not a general chatbot; it is tuned for hard decision problems, which makes it a better fit for workflows like risk assessment, resource allocation or strategic planning than for casual conversation.
- Multimodal and multilingual
The model handles multiple input types and supports 44 languages, so it can work across international teams and mixed-media data without separate translation layers.
- Benchmark leadership
It leads all eight categories of Liquid AI's 112-evaluation suite and scores 65 on the Decision Index, though the suite is internal and the index is the more independent signal.
- API-only access
There is no consumer app or dedicated page yet; you reach d1-ultra through an API, so you need development resources to use it.
How d1-ultra Works Under the Hood
Liquid AI has not published a full technical paper for d1-ultra, so the architecture details are limited. What is known is that the model is multimodal, meaning it can accept and reason over different data types, not just text. It supports 44 languages, which suggests a broad multilingual training corpus. The decision-model label implies the training and fine-tuning process emphasized tasks where the model must choose among options or produce a recommended course of action. Without official documentation, treat any specific claims about parameter count or training data as unconfirmed.
The Decision Index score of 65 is the most concrete external metric available. The Decision Index is a public benchmark that measures decision-making performance, and a score of 65 places d1-ultra at the top according to Liquid AI's announcement. However, benchmark scores can be sensitive to prompt formatting and evaluation conditions, so it is wise to run your own tests on tasks that match your use case. The claim of leading all eight categories of Liquid AI's 112-evaluation suite comes from Liquid's own testing, so keep that separate from third-party results when comparing models.
On pricing, the figure shared around launch is approximately $0.042 per million input tokens, with output tokens not billed. That is an unusual pricing structure; most APIs charge for both input and output. If accurate, it could make d1-ultra very cost-effective for workloads with long inputs and short outputs, such as classification or decision selection. However, pricing can change, so check Liquid AI's official pricing on liquid.ai before relying on that number for budgeting. The API-only access means you will need to handle authentication and rate limits yourself.
- Multimodal input
d1-ultra can process more than text, which opens up use cases like analyzing documents with images or mixed-media reports, though exact supported formats are not detailed in the announcement.
- 44-language support
The model handles 44 languages, making it suitable for global applications, but check the official site for the specific language list if your use case depends on a less common language.
- Decision Index score of 65
This public benchmark score is the clearest external validation, but benchmark performance does not always translate directly to real-world tasks, so validate with your own data.
- Unusual pricing model
At roughly $0.042 per million input tokens with output not billed, d1-ultra could be economical for input-heavy tasks, but confirm the current structure on liquid.ai.
Getting Started with the d1-ultra API
Because d1-ultra is API-only, your first step is to visit liquid.ai and look for developer documentation or a sign-up link. Liquid AI may require you to create an account and request API access. Once you have credentials, you can integrate the model into your application using standard HTTP requests. The API likely follows common patterns for multimodal models, but since there is no dedicated page yet, you should rely on the official documentation for exact endpoints and parameters. Do not assume compatibility with other providers' APIs without checking.
When you start testing, begin with a small set of decision tasks that reflect your actual use case. For example, if you are building a tool that recommends which support ticket to escalate, create a few sample inputs and see how d1-ultra ranks the options. Because the model is decision-focused, it may perform differently from general chatbots on open-ended prompts. Pay attention to how it handles ambiguity and whether it asks for clarification or makes assumptions. Document your prompts and results so you can compare against other models later.
Cost management is important even with low input pricing. Track your token usage, especially if you are sending large multimodal inputs. Since output tokens are reportedly not billed, you might be tempted to request long responses, but keep in mind that output quality and latency still matter. Set up monitoring and alerts to catch unexpected usage spikes. If you need higher throughput or dedicated support, ask Liquid AI about enterprise options. Finally, keep an eye on liquid.ai for a dedicated d1-ultra page, which may appear as the product matures.
- Check official documentation first
Since there is no dedicated page yet, liquid.ai is your source for API details, authentication methods and rate limits. Do not rely on third-party summaries for technical specs.
- Test with real decision tasks
Use prompts that mirror your production use case, such as ranking options or selecting a course of action, rather than generic chat prompts, to see the model's true strengths.
- Monitor token usage
Even at low input pricing, large multimodal inputs can add up. Set budgets and alerts, and remember that output tokens may not be billed but still consume time and compute.
- Compare against alternatives
Run the same tasks through other models you are considering, and compare not just accuracy but also latency, ease of integration and total cost.
How d1-ultra Compares and Who Should Use It
d1-ultra enters a market with many capable models, but its decision-model positioning is a differentiator. General-purpose models like those from OpenAI, Anthropic or Google are often used for decision tasks, but they are not specifically optimized for them. d1-ultra's focus could give it an edge on structured decision problems, especially if the Decision Index score holds up in independent testing. However, Liquid AI is a smaller player, so you may find fewer community resources, third-party integrations and tutorials compared to larger providers. That trade-off is worth considering if you need extensive ecosystem support.
The model's multimodal and multilingual capabilities make it attractive for global applications that involve mixed data types. If your decision problems involve analyzing images, documents or multilingual text, d1-ultra could reduce the need for separate preprocessing pipelines. The 44-language support is broad, though you should verify that your specific languages are included. On pricing, the quoted input rate is very low, which could be a significant advantage for high-volume, input-heavy workloads. However, the output-not-billed structure is unusual and may change, so do not build a business case solely on that detail without confirmation.
d1-ultra is best suited for developers and teams building decision-support systems, such as recommendation engines, risk assessment tools or automated planning assistants. It is less appropriate for casual users who want a chatbot, since there is no consumer app. If you are an individual researcher or hobbyist, the API-only access may be a barrier unless you are comfortable coding. For enterprises, the decision focus and multilingual support could justify a pilot. As with any new model, start with a small proof of concept, measure performance against your current solution, and scale only if the results justify it.
- Decision-focused vs general-purpose
d1-ultra is optimized for decision problems, while most competing models are generalists. If your tasks are decision-heavy, d1-ultra may outperform, but for open-ended chat, a general model may be more suitable.
- Smaller ecosystem
Liquid AI has a smaller developer community than major AI labs, so you may find fewer tutorials and third-party tools. Factor in the cost of self-support when evaluating.
- Multilingual and multimodal strengths
With 44 languages and multimodal input, d1-ultra can handle global, mixed-media decision tasks that might otherwise require multiple models or preprocessing steps.
- Pricing caveats
The low input price and free output are attractive, but they are launch figures. Confirm current pricing on liquid.ai and model your costs under different scenarios.
Step-by-Step Guide
Visit liquid.ai and look for developer access
Go to the official Liquid AI site at liquid.ai and search for API documentation or a developer sign-up. Since there is no dedicated d1-ultra page yet, you may need to explore the site or contact the company directly. Check for any announcements about availability.
Obtain API credentials
If API access is open, create an account and generate an API key. If it is gated, you may need to request access and wait for approval. Keep your credentials secure and never expose them in client-side code.
Review the API documentation
Read the official docs to understand endpoints, input formats, multimodal support and rate limits. Pay attention to how you specify languages and whether there are special parameters for decision tasks. Do not assume compatibility with other APIs.
Run a small pilot with your own data
Start with a few representative decision problems from your domain. Send them to d1-ultra and evaluate the responses for accuracy, consistency and usefulness. Compare against a baseline model you already use.
Monitor costs and scale gradually
Track your token usage and costs from the first day. Even with low input pricing, usage can grow. Once you see positive results, expand to more tasks and consider negotiating enterprise terms if you need higher volume.
Frequently Asked Questions
What is Liquid AI d1-ultra and who made it?
Liquid AI d1-ultra is a multimodal decision model released by Liquid AI on October 15, 2026. It is designed for hard decision problems rather than general conversation, supports 44 languages and scored 65 on the Decision Index. Liquid AI calls it its most powerful decision model to date.
How much does d1-ultra cost?
The launch figure quoted is approximately $0.042 per million input tokens, with output tokens not billed. That is an unusual structure, so check Liquid AI's official pricing on liquid.ai before relying on it. Prices can change, and enterprise plans may differ.
How can I access d1-ultra?
Access is offered through an API, so you need to visit liquid.ai to find developer documentation and sign-up details. There is no consumer app or dedicated product page yet. If API access is gated, you may need to request it and wait for approval.
What is the Decision Index score of 65 and does it matter?
The Decision Index is a public benchmark that measures decision-making performance. A score of 65 places d1-ultra at the top according to Liquid AI's announcement. However, benchmark scores can vary with testing conditions, so validate the model on your own tasks before making a decision.
Is d1-ultra better than general-purpose models like GPT or Claude?
It depends on your use case. d1-ultra is specifically optimized for decision problems, so it may outperform general models on structured choices. For open-ended chat or creative tasks, a general-purpose model might be more suitable. Always test both on your specific tasks.
