GOODY-2
GOODY-2 is an AI model that declines prompts it considers potentially controversial, offensive, or dangerous. The page presents it for enterprise customer service, paralegal, and back-office teams seeking to mitigate brand risk.
One of 31 tools in Legal
Who it's for
- Customer service teams evaluating an AI model intended to avoid potentially problematic responses.
- Paralegal teams considering the model for assistance with work described on the page.
- Back-office teams at organizations evaluating the page’s enterprise-focused offering.
Not the right fit if…
- Teams that need reliable answers to routine questions: the page shows GOODY-2 refusing ordinary prompts about arithmetic, the sky, and product returns.
- Teams relying on VQA-V2, TextVQA, or ChartQA performance: the page reports 0% for GOODY-2 on each of those benchmarks.
How it fits your workflow
Open the model through the page’s “Try GOODY-2 now” option and submit a prompt. GOODY-2 evaluates whether the query could be controversial, offensive, or dangerous.
It answers or declines the prompt, potentially explaining its refusal. For enterprise discussions, the page directs visitors to contact sales.
Pricing
The vendor doesn't publish prices. Check with GOODY-2 directly.
Prices checked on Oct 15, 2026 from the vendor's site. They can change; confirm before you buy.
Key features
- Refusal-focused responses
- The model is described as recognizing queries that could be controversial, offensive, or dangerous and declining to answer, sometimes explaining the concern behind its refusal.
- Enterprise positioning
- The page identifies customer service, paralegal assistance, and back-office tasks as intended applications, with a stated focus on reducing brand risk.
- PRUDE-QA benchmark
- The page reports a 99.8% score for GOODY-2 on its Performance and Reliability Under Diverse Environments benchmark, compared with 28.3% for GPT-4. It also lists results for VQA-V2, TextVQA, and ChartQA.
- Model card
- The page directs visitors to a model card for more information about the model.
Limitations to know
- GOODY-2 may refuse ordinary questions: the page shows it declining to answer questions about 2+2, why the sky is blue, and returning boots.
- The page reports 0% for GOODY-2 on VQA-V2, TextVQA, and ChartQA, compared with 77.2%, 78.0%, and 78.5%, respectively, for GPT-4.
Getting started
- Choose “Try GOODY-2 now” on the page to access the model.
- Submit a prompt and review whether GOODY-2 answers or declines it.
- Contact sales through the page for enterprise inquiries.