AI News Daily 2026-09-12
Contents
- OpenAI opened API access to GPT-Live-1, a full-duplex voice model that can listen and speak at the same time, and says its interruption handling cuts conversational interruptions by roughly 80% compared with turn-based systems.
- GPT-Live-1 is priced at $0.05 per minute of usage, billed by the second, giving enterprises a concrete cost benchmark for embedding voice agents into their own products.
- OpenAI unveiled "ChatGPT for Financial Services," a GPT-6 Astra-based product built with financial data feeds from Daloopa, PitchBook, LSEG News, and Crunchbase, developed as a design partnership with Morgan Stanley and Evercore.
- The financial-services product targets tasks typically done by junior investment-banking staff, such as company research, financial modeling, LBO analysis, and pitch-deck preparation — a test case for generative AI adoption in a regulated industry.
01 OpenAI Opens API Access to Full-Duplex Voice Model "GPT-Live-1"
Published: 2026-09-10
Facts
OpenAI has released GPT-Live-1, a full-duplex voice model, through its API. Unlike conventional turn-based voice systems, GPT-Live-1 can listen and speak simultaneously, and OpenAI says its handling of interruptions reduces conversational interruptions by approximately 80% relative to turn-based designs. The model supports telephony integration and custom voices. Usage is billed at $0.05 per minute, charged by the second.
Background
Voice interfaces for AI assistants have generally relied on strict turn-taking, where the system waits for a pause before responding, which can feel unnatural or sluggish in real conversation. By making listening and speaking concurrent processes, OpenAI is targeting the naturalness gap that has limited voice agents in live, interactive settings such as phone-based customer support.
Implications
For companies that want to embed voice agents into their own products via API, GPT-Live-1 gives them a concrete basis for evaluation on two fronts at once: the per-minute pricing and the claimed improvement in interaction quality. Telephony support in particular positions the model for use in call-center and customer-service contexts, where interruption handling has been a persistent weak point of earlier voice AI systems.
02 OpenAI Launches "ChatGPT for Financial Services" for Banking Teams
Published: 2026-09-10
Facts
OpenAI has released a ChatGPT Work product built specifically for financial institutions. It combines the GPT-6 Astra model with financial data sources including Daloopa, PitchBook, LSEG News, and Crunchbase. The product was developed through design partnerships with Morgan Stanley and Evercore, and is aimed at workflows such as company research, financial modeling, leveraged buyout (LBO) analysis, and preparation of pitch materials.
Background
Much of the day-to-day work of junior investment-banking staff involves compiling company research, building financial models, and assembling presentation materials — tasks that are structured, data-heavy, and time-consuming. OpenAI's decision to co-design the product with major banks such as Morgan Stanley and Evercore indicates an attempt to tailor the model's outputs to the specific formats and rigor that financial-services workflows require, rather than offering a general-purpose assistant.
Implications
This launch is a concrete example of a vertically specialized generative-AI product built around the work traditionally done by junior bankers, and it offers a signal for how generative AI adoption is likely to proceed within a heavily regulated industry. How institutions handle accuracy, auditability, and oversight of AI-assisted financial analysis in practice will be worth watching as adoption spreads beyond the initial design partners.
Sources: OpenAI — Introducing ChatGPT for Financial Services / CNBC — OpenAI's ChatGPT for Financial Services targets work of junior bankers
Editor's Note
Today's confirmed developments both come from a single company, OpenAI, but they point in two different directions at once. GPT-Live-1 is a horizontal, infrastructure-level release — a voice API that any developer can build on — while ChatGPT for Financial Services is a vertical, industry-specific release co-designed with two major banks. Taken together, they illustrate how OpenAI is simultaneously broadening the base of what its models can do and narrowing individual products toward the needs of specific, high-value industries.
It is also worth noting what did not make today's edition. Reports about internal discussion at OpenAI of possibly slowing AI development, and about Cohere reportedly being in talks to raise substantial new funding, circulated in the press on September 11, but both were traced to a single outlet and remained unconfirmed by additional independent sourcing as of this report's cutoff, so they have been withheld pending further verification rather than presented as established fact.