Doubleword
Share
Doubleword
Low-cost open-source model inference. Provides an OpenAI-compatible API for bulk tasks, batch processing, and agent execution.
General Information about Doubleword
Doubleword is a high-performance AI inference platform specifically designed to optimize the deployment of open-weight models. Its primary function is to provide an infrastructure that drastically reduces operational costs, allowing frontier-class models to run at a fraction of the price of traditional closed APIs. This tool is positioned as an advanced technical solution for teams managing high token volumes, complex data pipelines, and long-horizon AI agents.
Doubleword's technology is built on an inference engineering stack optimized at multiple levels to maximize efficiency. Key technical innovations include FlashOffload, which lowers the cost of prefill processes, and Speculative KV Coding, which enables up to four times higher cache compression. Additionally, its Cloudburst system accelerates cold starts, ensuring constant availability. The platform allows developers to choose from three service levels based on required latency: real-time, asynchronous, or 24-hour batch processing, enabling users to adjust spending based on the urgency of each task.
Integration is incredibly simple thanks to its OpenAI-compatible API. This allows engineers to migrate existing workloads by simply changing the base URL in their code, without needing to rewrite application logic. Doubleword supports a wide range of cutting-edge models such as Kimi-K3, DeepSeek-V4, Qwen, and GLM, covering needs ranging from logical reasoning and programming to multimodal processing and ETL data extraction.
Practical Capabilities and Benefits:
- Total cost control: Up to a 90% reduction in token spend through the use of open models and flexible service levels like the flex tier.
- Prompt caching: Speed and price optimization for repetitive tasks or extensive contexts, featuring high cache hit rates.
- Security and governance: Includes a Zero Data Retention (ZDR) policy, ensuring that processed information is not stored, meeting strict data security requirements.
- Scalability for agents: Infrastructure ready to support autonomous agent workflows that require complex reasoning and multiple recursive calls.
- Structured outputs: Native support for tool calling and specific output formats, facilitating interoperability with other software systems.
Doubleword is especially useful for companies performing large-scale data labeling, synthetic dataset generation, or those needing to deploy AI microservices with rigorous budget control. By focusing on inference efficiency, this tool unlocks use cases that were previously economically unfeasible with conventional providers.
Features and Use Cases of Doubleword
How Doubleword Works
Frequently Asked Questions about Doubleword
What is Doubleword and what benefits does it offer my business?
It is an inference platform optimized for open-weight models that allows you to run AI tasks with up to 90% cost savings.
Is it difficult to migrate my current applications to Doubleword?
Not at all, as our API is compatible with OpenAI and Anthropic standards, allowing you to simply change the base URL and get started in just a few minutes.
What service levels are available to control spending on Doubleword?
We offer three tiers based on task urgency: Real-Time for instant responses, Asynchronous for processes taking a few minutes, and Batch for 24-hour delivery.
Which models can I use within the Doubleword infrastructure?
You have access to a wide selection of leading open-weight models such as DeepSeek, Qwen, GLM, and Kimi, covering everything from simple tasks to advanced reasoning.
How does Doubleword guarantee the privacy of my input data?
We implement a standard zero-data retention policy to ensure that no information you send through our API is ever stored on our systems.
What is the Doubleword prompt caching feature?
It is a technology that stores frequent inputs so that repetitive queries are processed much faster and more affordably by eliminating the need to re-evaluate the entire text.
Can I request technical support to optimize my workflows on Doubleword?
Our engineering team provides direct assistance for system migration, prompt optimization, and large-scale processing queue adjustments.
Is there an option for companies with massive processing needs on Doubleword?
We offer a Scale plan that includes increased rate limits, custom models, regional data processing, and dedicated support through direct communication channels.
Doubleword Pricing
Self-Service Inference (Pay-as-you-go)
Pricing: Pay-as-you-go via prepaid credits. Rates vary by model and processing speed (Real-time, Asynchronous, or Batch).
- Access to all major open-weight models.
- Three priority tiers: Real-time (instant), Asynchronous (minutes/hours with a 25% discount), and Batch (24-hour processing with maximum savings).
- Includes prompt caching, tool calling, and structured outputs.
- Zero Data Retention (ZDR) included by default.
- Standard support and community access.
At-Scale Inference (High-Throughput)
Pricing: Custom volume-based rates (contact sales).
- Includes all Self-Service plan features.
- Increased rate limits.
- Custom models and inference optimizations tailored for high-volume workloads.
- Region-restricted data processing (U.S. or EU options).
- Direct engineering support, dedicated Slack channel, and Proof of Concept (PoC) assistance.
- Custom MSA and DPA contracts, and Service Level Agreements (SLAs).
Inference Rates by Model (Pricing per 1M tokens)
Costs are split into Input / Cache Read / Output. Pricing examples based on selected speed:
- Kimi K3: From $1.50 / $0.15 / $7.50 (Batch) to $3.00 / $0.30 / $15.00 (Real-time).
- GLM 5.3 Flash: From $0.08 / $0.02 / $0.25 (Batch) to $0.15 / $0.03 / $0.50 (Real-time).
- Qwen3.8 27B: From $0.25 / $0.02 / $1.50 (Batch) to $0.45 / $0.04 / $3.00 (Real-time).
- DeepSeek V4 Pro: From $0.65 / $0.07 / $1.30 (Batch) to $1.30 / $0.13 / $2.60 (Real-time).
- GPT OSS 20B: From $0.02 / $0.01 / $0.07 (Batch) to $0.03 / $0.02 / $0.13 (Real-time).
Doubleword Screenshots
