NEW YORK, Aug. 31, 2026 (GLOBE NEWSWIRE) — EdWealth (edwealth.ai) today announced the publication of MoneyBench, a benchmark developed by the company to evaluate how AI systems answer personal finance questions. The benchmark compares Ed, EdWealth’s AI personal finance coach, with ChatGPT and Gemini on usefulness and factual accuracy.
Ask an AI about money and a confident, plausible answer arrives instantly. Whether the numbers are right is invisible as you read. In an Intuit Credit Karma survey, 52% of adults who acted on AI financial advice reported a poor decision [1].
According to EdWealth’s MoneyBench results for July, Ed ranked first, winning 62.3% of questions, compared with 20.8% for Gemini and 17.0% for ChatGPT [2].
What made the difference
Writing quality did not separate the three, and accuracy scores often matched. Usefulness did: Ed averaged 4.24 on a five-point scale to ChatGPT’s 3.62 and Gemini’s 3.47, higher on roughly seven of ten questions against each.
Five things set Ed apart:
- Live numbers. Ed’s 120-plus financial data-and-analysis tools query the current filing, live price, holdings table, not last quarter’s memory.
- Decision-first answers. Conclusion first, then reasoning, then options.
- Fact and interpretation, separated. Limits and deadlines are stated as rules, context and trade-offs as interpretation. Most AI blurs the two.
- A coach that stays. From what a user shares, Ed learns their numbers, goals and habits; every answer builds on the last, across cash flow, taxes, stock compensation, funds, insurance. Account connections are read-only.
- Statute, not guesswork. Phase-outs, benefit formulas and multi-year tax rules come from a base of 11,566 parameters, sourced to the IRS, SSA, CMS and state authorities, across 51 US jurisdictions.
Confidence is not correctness
“Wealthy families always had a money person to call. Everyone else got search results, then a confident chatbot,” said Allen Ng, founder of EdWealth. “Ed closes that gap: live numbers, the reasoning shown, the decision still yours.”
The answers were scored by Claude, an AI from Anthropic, which builds none of the three. It checked each answer’s key figures against live sources, and an answer containing a fact proven false could not win. That check cost Ed: its score fell 5.3 points, both competitors rose, and Ed still finished first. Its 40 losses appear in the paper beside the wins. Where Ed’s facts were wrong, its usefulness fell with them: useful answers are built on correct ones.
In May’s first round Ed placed third of three, held back by weak data retrieval. EdWealth rebuilt it, and Ed has led every round since.
The part no benchmark can measure
The same question has a different right answer for each person: holdings, taxes, goals. A general assistant answers for everyone; Ed answers for one person, and keeps learning them. The test could not see that: standalone questions, nothing known about the asker. A benchmark measures the answer; a money person of your own knows the question behind it.
The full paper, with complete results, method and limitations: edwealth.ai/moneybench.
“We put our product on trial in public,” said Ng. “Financial AI should be judged on one question: does it help a person make a better money decision. Useful, and right. Only then does the rest follow. Money at peace, wealth in motion.”
About EdWealth
EdWealth builds agentic AI products for personal financial clarity and Financial Fitness. Its debut product, Ed, is a personal finance coach for modern households; user data is never for sale. Ed is available at edwealth.ai, on the App Store, and on Google Play.
Disclaimer: Ed provides financial information and education only — not investment, tax, or legal advice, and not a recommendation to buy or sell anything. Ed does not provide personalised investment recommendations. Ed is not a licensed financial adviser; its AI-generated outputs may be wrong, and all decisions are your own. Consult a licensed professional before acting. Availability, features, and pricing may vary by jurisdiction; Ed is offered only where permitted by applicable law.
Website: https://www.edwealth.ai/
Instagram: https://www.instagram.com/edwealth.ai/
Media contact: [email protected]
Sources: [1] Intuit Credit Karma, survey of 1,019 US adults, fielded August 7-14, 2025. [2] Systems as tested, July 2026: Ed in production configuration; ChatGPT (GPT-5.6 Sol) at Pro effort; Gemini (3.6 Flash) at default configuration.
Contact
Communications Lead
Phoebe Woo
EdWealth
[email protected]
Photos accompanying this announcement are available at:
https://www.globenewswire.com/NewsRoom/AttachmentNg/2435eb9d-52fc-4351-9db1-2c90ada70849
https://www.globenewswire.com/NewsRoom/AttachmentNg/ec7c40c0-0cba-416a-9caa-7ff13689cf41
https://www.globenewswire.com/NewsRoom/AttachmentNg/84c6e037-5072-4c88-a10a-9700bef81e34
https://www.globenewswire.com/NewsRoom/AttachmentNg/aedbbdd4-8503-4f03-84dd-46a5dd5b76c6
