Release Notes
GPT-5 Now Live for Assistant Q&A
Harvey Assistant just got smarter—GPT-5 is now the default model for all chat and draft responses, bringing major upgrades in reasoning, reliability, and writing quality.
What's New
We’re excited to announce that GPT-5 is now powering Harvey Assistant for chat style conversations. GPT-5 delivers a significant leap forward in intelligence, transparency, and output quality, helping you generate more accurate, well-structured, and natural responses.
- Transparent reasoning: GPT-5 now shows its “thinking” as it works, improving visibility and reducing perceived latency.
- More intelligence: Performs ~11% better than GPT-4.1 on Big Law Bench.
- Reduced hallucinations: Resolves numerous failure cases observed with GPT-4.1, offering more accurate and reliable outputs.
- Fixed table formatting: GPT-5 consistently produces valid markdown tables—resolving 100% of past GPT-4.1 table formatting issues in internal testing.
- Improved long-form writing: Generates fluent paragraphs instead of unwanted bullet points when longer narrative answers are requested.
Why it Matters
GPT-5 delivers sharper reasoning, clearer transparency, and higher-quality written outputs, empowering teams to produce better work faster and with greater confidence in accuracy.
How to Use
- Chat: Simply open any thread in Harvey Assistant and start a conversation—GPT-5 is now the default model for all chats.
- Draft: GPT-5 became the default for draft conversations in September, automatically ensuring improved structure and formatting.
- For more information on our models and FAQs, visit What AI Models Does Harvey Use?
Limitations
Availability: GPT-5 is now live for all Harvey Assistant users except those in Australia, where GPT-5 processing capabilities are still being developed. Access for AU customers will follow soon.