Doubleword

Doubleword is an AI inference platform that gives developers access to multiple open-weight models at a fraction of the cost of closed-source APIs. You can run models like Kimi-K3, DeepSeek-V4, and Qwen through three pricing tiers: realtime for instant responses, async for background tasks that complete in minutes to hours, and batch for jobs that can wait up to 24 hours. The platform claims up to 90% lower costs compared to providers like OpenAI or Anthropic, with average cache hit rates of 93% to reduce redundant compute. The headline comparison shows Doubleword's Kimi-K3 model scoring 59.7 on an intelligence benchmark while costing $18,000 per billion tokens, against GPT-5.6 Sol max at $24,000 and same score, or Claude Fable 5 at $60,000 for a slightly higher score.

The service works through a standard OpenAI-compatible API. You point your existing code at Doubleword's endpoint, pick a model, and optionally set the service tier to 'flex' for async or batch processing. The platform handles queuing, caching, and scaling behind the scenes. What makes Doubleword interesting is its focus on the 'efficient frontier' — the best cost-to-intelligence ratio at each performance level. Seven of the nine models on that frontier run on Doubleword, according to their data from Artificial Analysis. This is not a flashy consumer tool; it's a backend service for teams that need a lot of tokens processed cheaply.

The target audience is engineering and data science teams building AI-powered products at scale. Common use cases include running thousands of agentic evaluations, generating training datasets, annotating large image collections (as OpenMed did with 119K medical images), and powering high-throughput extraction or ETL pipelines. If your project needs frontier-level intelligence but the cost of closed APIs is eating your budget, Doubleword is worth a look. It is straightforward infrastructure — no magic, just cheaper inference for open models.