We Open-Sourced a 322M Decision Model. Now Its API Is Free

Описание и применение

HAL‑X AI выпустил открытый модель принятия решений THX‑01, работающую за ~10 мс и доступную через бесплатный API без ключа. Модель превосходит Jev 1.13 по точности и латентности, поддерживает 18 языков и может работать на смартфоне.

Задача

Developers need fast, low‑cost decision‑making models for tasks like spam detection, ticket routing, and numeric extraction, but large LLMs add unnecessary latency and cost.

Результат

THX‑01 outperforms Jev 1.13 with 98.4 % accuracy, 0.003 calibration error, and ~10 ms latency, while being 33× faster and fully open‑source.

Рабочий процесс

• 1. Send a JSON request with state and questions to https://api.hal-x.ai/v1/systemone
• 2. THX‑01 processes the input and returns structured answers with calibrated probabilities
• 3. Consume the API response in your application or pipeline

Доказательства

We benchmarked THX‑01 against TypeSafe Jev 1.13 using 2,843 multilingual support tickets across four languages.

Источник

Accuracy 98.4% vs 97.4%

Источник

Measured latency ~10 ms vs 331 ms

Источник

No API key, no sign‑up, and no payment required.

Источник