Wity
Wity is a decision-specific model introduced by Alphanimble that adopts the System 1.5 architecture, combining fast intuitive reasoning with on-demand deliberate reasoning and returning typed decision results with probabilities for each option, suited to high-frequency, fine-grained decision scenarios in software.
Timeline (last 90 days)
Wity is the first System 1.5 model, positioned as a model suited to high-frequency, fine-grained decision scenarios in software.
For most questions Wity completes one fast reasoning pass and gives an answer in about 0.1 seconds; when the answer is uncertain or unstable, it thinks before answering.
The Wity API accepts state (text, JSON, images or a combination) and one or more questions, with each question requiring listed options.
Wity offers four decision primitives: noul (yes/no), score (position on a scale), choice (pick one from options), and generate (generate text in a format).
Wity's returned probabilities let developers set confidence thresholds (e.g. auto-route only if probability exceeds 0.85, otherwise escalate to a human), and thresholds can be set per question.
Wity is multimodal, so photos, screenshots and scanned pages can be placed in the same request as text.
Wity supports a 128K-token context.
Wity supports on-premises deployment, with the API identical to the cloud version.
According to Wity's official blog, on the same 200 real cases compared against eight fast, low-cost models, a typical Wity call takes 109 milliseconds, while the second fastest takes 943 milliseconds.
According to Wity's official blog, in the same benchmark 1000 Wity runs cost $3.29, while the second cheapest costs $17.95.
3 more timeline events
All Facts (13)
Wity is the first System 1.5 model, positioned as a model suited to high-frequency, fine-grained decision scenarios in software.
50%UnverifiedFor most questions Wity completes one fast reasoning pass and gives an answer in about 0.1 seconds; when the answer is uncertain or unstable, it thinks before answering.
50%UnverifiedThe Wity API accepts state (text, JSON, images or a combination) and one or more questions, with each question requiring listed options.
50%UnverifiedWity offers four decision primitives: noul (yes/no), score (position on a scale), choice (pick one from options), and generate (generate text in a format).
50%UnverifiedWity's returned probabilities let developers set confidence thresholds (e.g. auto-route only if probability exceeds 0.85, otherwise escalate to a human), and thresholds can be set per question.
50%UnverifiedWity is multimodal, so photos, screenshots and scanned pages can be placed in the same request as text.
50%UnverifiedWity supports a 128K-token context.
50%UnverifiedWity supports on-premises deployment, with the API identical to the cloud version.
50%UnverifiedAccording to Wity's official blog, on the same 200 real cases compared against eight fast, low-cost models, a typical Wity call takes 109 milliseconds, while the second fastest takes 943 milliseconds.
50%UnverifiedAccording to Wity's official blog, in the same benchmark 1000 Wity runs cost $3.29, while the second cheapest costs $17.95.
50%UnverifiedWity is priced at $0.042 per million tokens, failed calls are free, and thinking is not billed.
50%UnverifiedWity gives new accounts $5 in free credit, to be used within 48 hours of signup.
50%UnverifiedWity does not try to replace LLMs in open-ended generation tasks, but rather solves problems that do not need long text and only need options with confidence scores.
50%