Monday, March 30, 2026

Google's new audio model + LLMs can't handle your money

Google just dropped Gemini 3.1 Flash Live with upgraded audio capabilities for conversational AI, while a new FinMCP-Bench benchmark shows LLMs still fumble complex financial reasoning (yikes). Meanwhile, Deccan AI closed a $25M Series A to tap India's talent pool for training infrastructure. Would you trust an AI agent with your portfolio yet?

Alibaba releases FinMCP-Bench, a comprehensive benchmark for evaluating LLM agents on real-world financial tool use with 613 samples across varying complexity levels. Evaluations show even leading models like Qwen3-235B struggle with complex multi-tool reasoning and multi-turn conversations, highlighting significant gaps in agentic financial AI capabilities.

agentsllmbenchmarkfinancial-ai

Gemini 3.1 Flash Live

Google launched Gemini 3.1 Flash Live, a new audio AI model with improved naturalness, lower latency, and stronger reasoning capabilities, now powering Gemini Live, Search Live globally, and available to developers and enterprises. The model leads on key benchmarks and includes SynthID watermarking for AI-generated audio detection.

geminigoogleaudio-aillm

Deccan AI — $25M Series A

TechCrunch

Deccan AI raised $25M Series A to provide AI post-training services through an India-concentrated workforce of over 1 million contributors, serving clients like Google DeepMind. The startup's approach emphasizes quality control by focusing operations in one geography, addressing the growing demand for high-skill AI training work as models become more complex.

fundingai-trainingindiaenterprise-ai