{"slug": "we-open-sourced-a-322m-decision-model-now-its-api-is-free", "title": "We Open-Sourced a 322M Decision Model. Now Its API Is Free", "summary": "HAL-X AI has open-sourced THX-01, a 322M-parameter multilingual decision model released under Apache 2.0 with weights, source code, and training scripts, and has made its hosted API free with no key or sign-up required. On a benchmark of 2,843 multilingual support tickets across four languages, THX-01 scored 98.4% accuracy with a 0.003 calibration error and roughly 10 ms latency, versus 97.4%, 0.007, and 331 ms for TypeSafe Jev 1.13, though the company notes the latency figures come from different serving setups and are not a controlled hardware comparison.", "body_md": "**Two days ago, we released THX-01, an open-source decision model that outperforms Jev on several benchmarks and can even run locally on a smartphone. Today, anyone can use it through our free API.**\n\nHi everyone! I'm Farid, Co-Founder of HAL-X AI.\n\nA couple of days ago, our team open-sourced **THX-01**, a 322M-parameter multilingual decision model designed to make AI decision-making faster, cheaper, and more accessible.\n\nYesterday, we opened our hosted API to everyone. **No API key, no sign-up, and no payment required.**\n\nAnd the response has been incredible!\n\nSince launch, we've already processed:\n\nWe're excited to see developers experimenting with what a small, specialized decision model can do.\n\nMost applications don't actually need an LLM to generate paragraphs of text.\n\nSometimes, you just need to answer a simple question:\n\nUsing a large generative model for these tasks can introduce unnecessary latency and cost.\n\n**THX-01 takes a different approach.**\n\nIt's a non-autoregressive model that processes a document or message alongside typed questions and returns structured answers with calibrated probabilities in a single forward pass.\n\nApproximately **10 milliseconds** on our benchmark hardware.\n\nNo token-by-token generation.\n\nYou can make your first request right now.\n\nNo authentication required.\n\n```\ncurl https://api.hal-x.ai/v1/systemone \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"state\": \"My card was charged twice for one order\",\n    \"questions\": {\n      \"team\": {\n        \"type\": \"choice\",\n        \"question\": \"Which team handles this?\",\n        \"criteria\": {\n          \"billing\": \"billing\",\n          \"tech\": \"technical\",\n          \"sales\": \"sales\"\n        }\n      }\n    }\n  }'\n```\n\nTHX-01 evaluates the available choices and returns the predicted decision with probabilities.\n\nThe API supports:\n\n| Type | What it does | \n|---|---|\n| Choice | Classification with probabilities | \n| Yes/No | Binary decisions | \n| Score | Ordered scoring | \n| Number | Extract exact numbers from documents | \n| Excerpt | Extract verbatim passages | \n| Citations | Identify supporting text | \n\nYou can submit multiple questions in a single request.\n\n**Full API documentation:** [https://api.hal-x.ai/docs/thx-01](https://api.hal-x.ai/docs/thx-01)\n\nWe benchmarked THX-01 against TypeSafe Jev 1.13 using 2,843 multilingual support tickets across four languages.\n\nHere are the results:\n\n| Metric | THX-01 | Jev 1.13 | \n|---|---|---|\n| Accuracy | **98.4%** | 97.4% | \n| Calibration error | **0.003** | 0.007 | \n| Measured latency | **~10 ms** | 331 ms | \n| Open weights | Yes | No | \n\nTHX-01 achieved higher accuracy and better calibration on this benchmark, with approximately 33x lower measured latency.\n\nThe latency measurements use different serving setups, so they're not a controlled hardware-to-hardware comparison.\n\nWe don't claim THX-01 outperforms Jev everywhere, but these results demonstrate how competitive specialized, lightweight decision models can be.\n\nTHX-01 contains just **322 million parameters**.\n\nIt was designed to support efficient inference without requiring a massive GPU infrastructure.\n\nWith a suitable runtime, it can even run locally on smartphones.\n\nYou can also install the Python package:\n\n```\npip install thx01\n```\n\nThe model supports 18 post-training languages, with particular attention to Azerbaijani and multilingual robustness.\n\nWe're releasing THX-01 under **Apache 2.0**, including its model weights, source code, and training scripts.\n\nWe believe developers should have the freedom to run, modify, deploy, and build on top of their AI models.\n\nAt HAL-X AI, our goal is to contribute to the global open-source AI ecosystem and make efficient AI accessible to more people.\n\nThat's also why we've made the hosted API free to use, subject to rate limits.\n\n**Free API:** [https://api.hal-x.ai/docs/thx-01](https://api.hal-x.ai/docs/thx-01)\n\n**Model weights and code:** [https://huggingface.co/doofz/THX-01](https://huggingface.co/doofz/THX-01)\n\n**Interactive demo:** [https://huggingface.co/spaces/doofz/THX-01-demo](https://huggingface.co/spaces/doofz/THX-01-demo)\n\n**Python package:**\n\n```\npip install thx01\n```\n\nWe're still improving THX-01, and we'd love to hear feedback from developers building AI agents, automation pipelines, document processing systems, and classification services.\n\nIf you build something using THX-01, we'd love to see it!", "url": "https://wpnews.pro/news/we-open-sourced-a-322m-decision-model-now-its-api-is-free", "canonical_source": "https://dev.to/montydrief/we-open-sourced-a-322m-decision-model-now-its-api-is-free-187e", "published_at": "2026-10-10 21:34:17+00:00", "updated_at": "2026-10-10 21:46:17.502071+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-products", "ai-tools", "ai-startups"], "entities": ["HAL-X AI", "THX-01", "TypeSafe Jev 1.13", "Farid", "Hugging Face"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/we-open-sourced-a-322m-decision-model-now-its-api-is-free", "markdown": "https://wpnews.pro/news/we-open-sourced-a-322m-decision-model-now-its-api-is-free.md", "text": "https://wpnews.pro/news/we-open-sourced-a-322m-decision-model-now-its-api-is-free.txt", "jsonld": "https://wpnews.pro/news/we-open-sourced-a-322m-decision-model-now-its-api-is-free.jsonld"}}