GPT-5.5: 52% Accuracy in Finance, Can't Replace Analysts

The most advanced AI model, GPT-5.5, achieves only 52% accuracy on real-world financial analyst tasks. This is according to the latest Vals AI Finance Agent v2 benchmark. The benchmark, released in May 2026, shows that AI cannot yet replace human financial analysts. Large language models still fail about half of the multi-step research, modeling, and data-retrieval tasks that junior analysts perform daily. The Vals AI Finance Agent v2 tests models on realistic financial analyst workflows, not just simple questions. It includes tasks like extracting figures from filings and building financial inputs. GPT-5.5 was the top performer in this May 2026 evaluation, scoring around 52% accuracy. Other frontier models like Anthropic's Claude and Google's Gemini also scored in the high 40% to low 50% range. This 52% score, while modest, represents progress from earlier models that scored in the 30-40% range in 2024. This information is important for traders, investors, and crypto market participants who use AI for research.