OpenAI Says Basis Did a Tax Workbook in Half the Time
OpenAI published a customer story about Basis, a startup building AI agents that automate much of the manual work accountants do each day. Everything below is OpenAI describing results its own customer reported.
Transcript
OpenAI published a customer story about a startup using GPT six Astra. Its numbers are self reported.
Basis builds AI agents that automate manual accounting work. OpenAI says Basis sees stronger understanding of intent.
Basis tested GPT six Astra against the earlier model Sol on a fifty tab tax workbook. Basis says Astra finished in half the time.
Basis says Astra makes better early decisions and uses more reasoning only on hard steps. Basis reported about twenty percent better internal evaluation scores.
This is OpenAI's marketing page about its own customer. The numbers come from Basis's own testing with no methodology or independent verification.
Inboxsmith builds AI receptionists for small business. Please like and subscribe for more news.
Sources
Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:
- Basis builds AI agents to automate much of the manual work that accountants do each day, helping them shift their time from repetitive tasks to strategic work.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- The company's research focuses on agents that can reliably complete long tasks, and with GPT-6 Astra, it's seeing a stronger understanding of what accountants want to accomplish.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- Basis compared GPT-6 Astra and GPT-5.6 Sol on a complicated tax workbook with 50 tabs. The task was to complete the workbook accurately and reliably, and GPT-6 Astra was markedly faster.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- "GPT-6 Astra is able to complete that workbook in half the time that GPT-5.6 Sol is able to."(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- Basis also noted that GPT-6 Astra makes better decisions at the start of a task, helping Basis's agents take a more direct path through the work with less time spent correcting mistakes.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- Basis also has GPT-6 Astra adjust how much reasoning it uses as a task progresses, dialing up computation when a step is difficult, and using less when a step is easier.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- Basis saw about a 20% improvement in its internal evaluation scores with GPT-6 Astra, driven by better understanding of user intent, including when to ask questions, flag assumptions, and follow instructions.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
- Verified absence: the page carries no independent or third party verification of any number, and both figures come from Basis's own internal testing as published by OpenAI. It publishes no methodology, so it does not say how many runs were done, what the accuracy bar was, what the workbook contained, what settings or reasoning levels were used, or who supervised the comparison. It contains no pricing, no availability change, no new feature and no new model, because GPT-6 Astra already exists and this is a customer story about using it. It gives no absolute timings, no cost figures and no token counts, and it reports no customer count, revenue figure or adoption number for Basis. The page is OpenAI authored marketing in the company's customer stories template, with Company size, Region, Industry and Products fields, rather than a news report or a benchmark publication. The YouTube transcript pull for this story returned zero of five transcripts, so no video corroboration existed and every narration claim traces to this page alone.(OpenAI (openai.com), first party customer story titled "Basis completes a tax workbook 2x faster with GPT-6 Astra", dated 2026.09.28, live URL returned HTTP 403 so the alert time snapshot captured 2026-09-28T21:16:47.717Z was used as primary.)
