# How OpenAI Built ChatGPT for Bankers and Analysts — Joseph Kim, OpenAI

Joseph Kim, OpenAI

[Talk page](https://ai.engineer/talks/opHmX7qClzE-openai-built-chatgpt-bankers-analysts) · [Video](https://www.youtube.com/watch?v=opHmX7qClzE)

## How OpenAI Built ChatGPT for Bankers and Analysts

OpenAI kept MCP while taking control of hosting and search for financial data, then added firm templates, interactive charts and citation review. Joseph Kim connects those decisions to the gap between model capability and useful work—and to his bets for finance in 2027.

![Recording frame at 4:54: How OpenAI Built ChatGPT for Bankers and Analysts](https://ai.engineer/api/talk-essay-media/245d541ae827a074cd7982352d5d2f4aef47b187ac53a4e17feeea04e5c475ad)

How OpenAI Built ChatGPT for Bankers and Analysts — recording at 4:54

### Financial work needs sources, review and the firm’s format

[0:12](https://www.youtube.com/watch?v=opHmX7qClzE&t=12s)

Software engineers were already using tools such as Codex to change their daily work, while many financial professionals had yet to feel a comparable shift. Joseph Kim, OpenAI’s lead for ChatGPT for Financial Services, starts with that discrepancy: model intelligence was improving quickly, but turning it into useful financial work still required products that fit people’s daily responsibilities.

![Recording frame at 2:29: Financial work needs sources, review and the firm’s format](https://ai.engineer/api/talk-essay-media/77d751bcd306e725e06ba1b961d6e3199c6a349fd1d3079fe24b7ac12895d049)

Financial work needs sources, review and the firm’s format — recording at 2:29

ChatGPT for Financial Services, launched a month before the talk, is a tailored version of ChatGPT Work. Its initial focus is investment banking and equity research at large financial institutions. OpenAI developed it with design partners Morgan Stanley and Evercore, concentrating on the requirements that surround an analyst’s work as well as the model’s ability to answer questions.

The initial design addresses three practical needs:

- **Financial sources:** Give people access to the information their work depends on, directly within ChatGPT. The quality of an analysis is limited by the quality of its sources.

- **An interface for investigation and review:** Let analysts inspect source material, explore financial charts and control which sources inform an analysis.

- **The firm’s format:** Help generated models and presentations follow the organization’s formats and standards through firm-specific templates. A deliverable has to work for the team that will review and share it.

### The Codex adoption curve and the capability overhang

[4:11](https://www.youtube.com/watch?v=opHmX7qClzE&t=251s)

The distance between what a model can do and the value people obtain from it is the **capability overhang**. A model can improve substantially while a professional’s working day barely changes. Closing that gap requires usable access to data and tools, together with an experience that fits the work. This gives product design a direct role in making existing intelligence useful.

An internal Codex usage chart addresses a concrete question: when did colleagues outside engineering begin putting the tool to work? From August 2025 to June 2026, the chart tracks, by job function, the average user’s share of monthly output tokens generated in Codex. Engineers started early and their share grew steadily. In finance, legal and recruiting, that share was near zero at the beginning of 2026 and around 90% by June. Kim points to a pronounced upturn around February and March.

![Recording frame at 5:29: The Codex adoption curve and the capability overhang](https://ai.engineer/api/talk-essay-media/edb45d01434f6e0be9b5bae6fa673e4e4a49124b0e5ae8a73b6e2cdcc31e7220)

The Codex adoption curve and the capability overhang — recording at 5:29

Kim initially went looking for a model release to explain that turn. What he found was the February release of the Codex desktop app. His interpretation is that a more accessible product helped people outside engineering use capabilities already available to them. The timing alone does not isolate the app’s causal effect, but it supplies a concrete reason to take interface changes seriously. The finance product pursues the same opportunity: make capable models easier to put to work in a particular profession.

### License and index the data while keeping MCP

[6:01](https://www.youtube.com/watch?v=opHmX7qClzE&t=361s)

The data foundation had to cover sources professionals already rely on: financial statements, earnings transcripts, market data and private company data. Existing MCP connectors offered a way to reach that information, but many of the connectors the team examined had problems with latency, errors or search quality. A model’s ability to call a tool did not resolve those problems in the information it received.

![Recording frame at 6:34: License and index the data while keeping MCP](https://ai.engineer/api/talk-essay-media/13a1d42b4a29a785abdb2659bc402af79a69abbc444574ee962d380870f585f4)

License and index the data while keeping MCP — recording at 6:34

OpenAI licensed much of the data directly and built its own hosting and search indices using the same infrastructure as its web index. MCP remained the protocol through which the model accessed the data. The change was ownership of the infrastructure behind that access. Taking on licensing, hosting and indexing gave the team more control over search quality and the reliability of the experience, while also making it responsible for those parts of the product.

That investment also creates an option for post-training models to work with the datasets. MCP and skills provide access and instructions; post-training can develop familiarity with the data itself. Kim’s analogy is a junior analyst with an instruction manual compared with a senior analyst who has spent more than two decades learning a dataset’s quirks. The useful distinction is between being told how to reach and use information and having learned its peculiarities through training.

How does one data investment support retrieval and training?
The runtime branch supplies financial information through MCP. The separate post-training branch represents an option for improving familiarity with the datasets. These are different uses of the same data foundation; training is not a step in serving a live retrieval request.
- Directly licensed financial data — The team licensed much of the information it wanted to provide.
- OpenAI hosting and search indices — Built using the same infrastructure as the web index.
- MCP access — Makes financial information available during the model’s work.
- Dataset-specific post-training — An option for teaching familiarity with the data and its quirks.
- Model doing financial work — Uses retrieved information; post-training can improve how it works with the datasets.
- Directly licensed financial data → OpenAI hosting and search indices (Host and index the data)
- OpenAI hosting and search indices → MCP access (Make financial information accessible)
- MCP access → Model doing financial work (Supply information at runtime)
- OpenAI hosting and search indices → Dataset-specific post-training (Enable training with these datasets)
- Dataset-specific post-training → Model doing financial work (Can improve dataset familiarity)

### Firm templates and source review on the plugin platform

[7:01](https://www.youtube.com/watch?v=opHmX7qClzE&t=421s)

Financial work often ends in an Excel model, a PowerPoint presentation or a written analysis that other people must review and share. The template system makes the firm’s approved templates available when generating spreadsheets and presentations. Its intended saving is the preparation work between obtaining an output and making it ready for a team or client.

Interactive financial charts support the investigation that happens along the way. Analysts can explore stock prices and financial metrics directly inside ChatGPT while working through an analysis. The interface therefore supports examining the information as well as producing a final document from it.

For a financial figure, the review question is concrete: where did this number come from, and what context surrounds it? Citation viewers let people inspect the source passages and tables behind an answer. That connects the generated analysis to material the analyst can examine.

![Recording frame at 8:32: Firm templates and source review on the plugin platform](https://ai.engineer/api/talk-essay-media/8e9af577f7903c3111446b306d7123cd42428d6648114d10d116156255462e26)

Firm templates and source review on the plugin platform — recording at 8:32

The connection makes source review a normal part of working through an answer. The reader can examine whether the cited material supports the figure and its interpretation. Much of this experience uses the same ChatGPT plugin platform offered to developers, with different components serving different jobs:

- **Data access:** MCP tools and connectors give the model the financial information it needs.

- **Source inspection and charts:** Citation metadata and MCP apps support inspecting sources. MCP apps also power the interactive financial charts.

- **Workflows and deliverables:** Skills work with shared artifact tools to guide financial workflows and produce Excel models and PowerPoint presentations.

- **Onboarding:** The product uses the platform’s native plugin onboarding experience for financial onboarding.

For builders, the valuable contribution can be unique data, an existing tool or a deep understanding of a particular workflow. The platform supplies building blocks around which that expertise can become a usable product. Kim is bullish enough on this approach to predict multiple billion-dollar businesses built as ChatGPT plugins, with the opportunity expanding as models become more capable.

### Why Kim expects the capability curve to keep moving

[9:18](https://www.youtube.com/watch?v=opHmX7qClzE&t=558s)

The forward-looking half of the talk begins with three overlapping drivers of model improvement: scaling compute, improving algorithms and **unhobbling**, the changes that let people get more useful work out of models. Looking backward at those drivers is Kim’s way of building an intuition for what might become possible next.

![Recording frame at 10:04: Why Kim expects the capability curve to keep moving](https://ai.engineer/api/talk-essay-media/cd9d0c83182a71d73962893b5796777aafe0bcc86fa7ed2105ef4abce002083b)

Why Kim expects the capability curve to keep moving — recording at 10:04

Pre-training scaled capability by making models larger and adding more data. Before **GPT-4** finished training, OpenAI researchers could predict its final loss and its performance on a coding evaluation. Algorithmic improvements supplied another kind of gain: Kim cites a study in which the compute required to train a language model from scratch to a fixed performance level fell by roughly half every eight months. That is an efficiency trend at a given target, even though the timing of individual breakthroughs remains hard to predict.

Reasoning added another route in 2024 and 2025. Reinforcement learning and post-training taught models to work through problems more effectively, while additional compute at answer time gave them more opportunity to think. These changes operated alongside pre-training. Kim also reports that pre-training problems had made scaling difficult at OpenAI, and that resolving many of those problems in 2026 enabled larger, more capable models.

Potential barriers had prompted concern that this progress might stop. The feared data wall was a shortage of human-generated training text. In Kim’s assessment, synthetic data generation, learning from verifiable rewards and learning more from each token had largely addressed that constraint. He similarly points to sparse attention and compaction as answers to context-length constraints. His expectation is continued progress toward larger, more efficient and more capable models in 2027, while alignment and safety remain difficult problems.

Mathematics supplies the concrete trajectory: GPT-4 scored 700 out of 800 on SAT math in 2023, followed later by an OpenAI model reaching gold-medal-level performance at the International Mathematical Olympiad. Kim then reports internal research work: disproving a long-standing conjecture connected to a problem studied since 1946, proposing a solution to the Navier–Stokes Millennium Prize Problem, and sharing 719 mathematical research manuscripts across 372 families of results. These later examples are research claims; a proposed solution and a collection of manuscripts do not by themselves establish accepted mathematical results.

The extrapolation is conditional on another year of progress roughly in line with the recent pace. Under that condition, Kim expects models by the end of 2027 to outperform leading human experts on a growing range of difficult tasks across fields and specialties. Mathematics is his example of how a capability can move from familiar tests toward work at a field’s frontier.

### The threshold is whether checking the work still saves time

[13:13](https://www.youtube.com/watch?v=opHmX7qClzE&t=793s)

A gradual capability curve can produce a sudden change in working habits. Kim calls this the **threshold of usefulness**: the point at which a model becomes capable and reliable enough to save time after the user has checked and corrected its work. The threshold belongs to a particular task and depends on how much review that task requires.

![Recording frame at 14:09: The threshold is whether checking the work still saves time](https://ai.engineer/api/talk-essay-media/6e1f3e1051de1882d280e05df73ae29855e8ef67e8a495349e815f86dc94b301)

The threshold is whether checking the work still saves time — recording at 14:09

What changes on either side of the threshold is the practical value of the same task:

| Aspect of the task | Below the threshold | Above the threshold |
| --- | --- | --- |
| Output | Answers may improve and contain fewer mistakes, but still require substantial correction. | The model produces something the user can use. |
| Human effort | Checking and fixing take as long as doing the task oneself. | Review takes much less time than the work the model saves. |
| Working habit | Capability gains can feel like little practical progress. | Regular use becomes worthwhile, followed by greater comfort with delegation. |

Several similar increments in capability can therefore feel very different to the user. Early improvements reduce errors without changing the decision to do the work manually. A later improvement makes the output usable with a manageable review cost, and the decision changes. The underlying gain need not be unusually large for the change in behavior to feel dramatic. Different tasks have different thresholds, so these transitions will arrive at different times.

This creates product work that can begin before the next capability gain arrives. Identify a task whose model performance is easy to underestimate because it is almost useful, then consider the experience needed once delegation becomes worthwhile. How should the interface change? What should the user still control, review or approve? Having those decisions worked out helps a new capability become useful as soon as it crosses the threshold.

### A rising tide of capability, with finance-specific waves

[14:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=883s)

Financial tasks have two routes toward that threshold. Kim describes them as a rising tide and the waves on top:

- **The rising tide:** Larger, more capable general models bring transferable skills from mathematics, coding and reasoning into finance. Historically, OpenAI had devoted relatively little training time or compute specifically to finance, so much of the financial capability came from those broader improvements.

- **The waves:** Finance-specific post-training builds on the general model and concentrates learning on the work financial professionals do. It can create stronger performance on particular tasks above the improving general baseline.

![Recording frame at 15:54: A rising tide of capability, with finance-specific waves](https://ai.engineer/api/talk-essay-media/267d315a29c5e1e98c77b11379d902d818b2630ef87abbb0ed4fb55257ac687f)

A rising tide of capability, with finance-specific waves — recording at 15:54

At the time of the talk, the team’s next three months of work focused on general financial reasoning and the quality of common analyst outputs: analyzing financial information, building Excel models and creating PowerPoint presentations. These tasks occupy a significant part of a junior analyst’s day. The training objective is work that is accurate, well structured and useful within an actual workflow.

The longer-term target moves into the judgment of senior bankers, analysts and investors. That work involves taking in large amounts of information, deciding what matters and making forward-looking assessments: forecasting a company’s earnings, assessing where commodity prices may go or evaluating which businesses make attractive investments. A model has to select the important information and use it to reason about an uncertain future.

Kim’s stated goal is to match the best human experts on these financial tasks sometime in 2027. He expects both forces to contribute: continued improvement in the underlying model and more direct training on financial work. The broad model supplies transferable capability; the specialized training concentrates it on the tasks the product needs to perform.

### Persistence changes the unit of interaction to a project

[16:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=1003s)

Unhobbling returns as the third driver of useful progress. Earlier examples include RLHF helping models follow instructions and hold conversations, and tools letting them search for information, write and run code, and interact with software. Each change gives model capability a more useful way to operate in the world.

![Recording frame at 17:39: Persistence changes the unit of interaction to a project](https://ai.engineer/api/talk-essay-media/3f887fa0b553f5a10ec95176f562d01b1641521057b0523e15480a25b57c0478)

Persistence changes the unit of interaction to a project — recording at 17:39

The next form Kim expects is a proactive, persistent agent. A persistent agent would retain a goal, remember relevant context and continue working in the background after receiving direction. Progress would no longer depend on the user being present to supply the next prompt. The assignment could remain active while the person turns to other work.

Persistence also changes how people cooperate with the agent. Getting useful work from today’s agents often requires learning how to prompt them and operate their tools. With continuity across an assignment, the interaction can become more familiar: give the agent a project, share context and ask for an update. The accessibility benefit comes from organizing the relationship around ongoing work, much as people coordinate with colleagues.

### Three bets for finance in 2027

[17:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=1063s)

Kim closes with three explicit predictions for 2027:

- **Less reliance on Excel and PowerPoint to make the work:** People will rarely, if ever, rely on those applications to create financial models or pitch decks.

- **AI analysts that behave more like coworkers:** They will remember preferences and context from previous days, take ownership of assignments, proactively follow up and keep working while the person focuses elsewhere.

- **Better forward-looking judgments than leading human experts:** Models will begin to outperform the best people on critical financial tasks, including forecasting earnings and developing investment ideas.

![Recording frame at 18:49: Three bets for finance in 2027](https://ai.engineer/api/talk-essay-media/8dc111e23071fbc48a2b0c9daeb9b6e577156b271938690774123c7bf0e69f2e)

Three bets for finance in 2027 — recording at 18:49

The product already aims to produce Excel models and presentations in a firm’s format. These bets extend that direction into how assignments are managed and the quality of the judgments inside the deliverables. The design question is what a user will still need to control, review and approve when delegation becomes worthwhile—and what builders should start making now so that experience is ready.

## Key takeaways

- The capability overhang is the gap between model capability and useful daily work. ChatGPT for Financial Services addresses that gap through financial data, an interface for investigation and review, and outputs that follow the firm’s format.
- Keeping MCP is compatible with owning the retrieval infrastructure. OpenAI licensed and indexed financial data to gain more control over quality and reliability, while preserving MCP access and creating an option for dataset-specific post-training.
- The plugin platform supplies distinct building blocks: MCP tools for data access, citation metadata and MCP apps for source inspection, MCP apps for charts, and skills with artifact tools for financial workflows and deliverables.
- A task crosses the threshold of usefulness when checking and correcting the model’s work costs substantially less time than the work it saves. Similar capability improvements can therefore produce very different changes in adoption and delegation.
- The finance training roadmap combines general improvements with specialized post-training. The near-term focus is analyst reasoning and usable Excel and PowerPoint outputs; the longer-term goal is expert-level judgment about earnings, commodity prices and investments.
- The proposed persistent agent retains a goal and relevant context while continuing in the background. That changes the interaction from repeatedly directing individual steps toward assigning a project, sharing context and checking progress.

## Related talks

- [How Kepler Built Verifiable AI for Financial Services](https://ai.engineer/talks/Tt2kX2sgQio-kepler-built-verifiable-ai-financial-services) — Develops the verification side of financial AI by separating model reasoning from deterministic retrieval, computation and source-level provenance, extending the question of what an analyst needs to inspect before trusting a work product.
- [Challenges to Scaling Agents for Generative AI Products](https://ai.engineer/talks/b2GqTDWtg6s-challenges-scaling-agents-generative-ai-products) — Adds Bloomberg’s perspective on structured and unstructured financial data, search-index evolution, factuality requirements and errors that compound across composed agents.
- [Claude for long-horizon tasks](https://ai.engineer/talks/9QebvrrY3KY-claude-long-horizon-tasks) — Adds architectural detail to the persistent-agent direction through externalized session context, separate-context verification and memory consolidation for work that continues over longer periods.

## Slides

### Building the Next Chapter of AI in Finance

Joseph Kim introduces ChatGPT for Financial Services, launched one month before the talk. He focuses on turning model intelligence into products people in finance can use every day.

[0:17](https://www.youtube.com/watch?v=opHmX7qClzE&t=17s) [OpenAI title slide reading Building the Next Chapter of AI in Finance, with Joseph Kim identified as Head of ChatGPT for Financial Services at lower left. A green bank symbol appears in a white rounded square on a blue and green gradient panel at right.](https://assets.ai.engineer/sites/aiecode2025/sha256/85851fe102cf75ceb3508ff03f0f255c2b8afd8176fd779791a517799eea1122.webp)

Building the Next
Chapter of AI in
Finance

OpenAI

Joseph Kim
Head of ChatGPT for Financial Services

1

### Talk agenda

Kim will demo ChatGPT for Financial Services, explain why it was built and share technical lessons, then explore what changes in models and products could mean for financial work over the next year.

[1:02](https://www.youtube.com/watch?v=opHmX7qClzE&t=62s) [White agenda slide listing four numbered sections: 01 What we’ve built, 02 Why we built it, 03 How we built it, and 04 What comes next. OpenAI appears at bottom left and page number 2 at bottom right.](https://assets.ai.engineer/sites/aiecode2025/sha256/3d3ccf8d6dfed39d048c8c5c1e3e140a48e70a5f571e5e6e528848d3f4766622.webp)

01 What we’ve built
02 Why we built it
03 How we built it
04 What comes next

OpenAI
2

### 01 What we’ve built

Joseph Kim opens the first section of the talk: a walkthrough of ChatGPT for Financial Services, with a quick demo to follow.

[1:24](https://www.youtube.com/watch?v=opHmX7qClzE&t=84s) [Section 01, What we’ve built, appears in a large white rounded panel on an aqua and green background. OpenAI and page number 3 appear along the slide's bottom edge. A partially cropped event banner beneath the slide repeats the talk title and presenter.](https://assets.ai.engineer/sites/aiecode2025/sha256/1278118a18995c2f16d1a9af8cc8c8b19dcef5328fa3c922957553d62cfd6084.webp)

01
What we’ve built

OpenAI
3

Building the Next Chapter of AI in Finance
Joseph Kim / Head of ChatGPT for Financial Services
OpenAI

### ChatGPT for Financial Services

The finance version of ChatGPT Work pairs premium data with tools to inspect and control sources. Firm templates guide users’ outputs, with ChatGPT Enterprise security and controls underpinning all three areas.

[1:28](https://www.youtube.com/watch?v=opHmX7qClzE&t=88s) [Slide titled ChatGPT for Financial Services with three illustrated columns covering financial data, an interface built for finance, and firm-specific templates and artifact generation. The footer states that it is built on ChatGPT Enterprise security and controls.](https://assets.ai.engineer/sites/aiecode2025/sha256/7c0c475ad944bf285bff5e5e300c6c149628723bbc4ef62d243339fc11909d74.webp)

ChatGPT for Financial Services

A tailored version of ChatGPT Work built for investment banking, equity research, and more.

Financial data
Built-in premium data, automatic data entitlements, and improved MCP connectors.

UI Built for finance
Inspect source passages and tables, explore interactive charts, and control the sources behind your analysis.

Your firm’s format
Centrally managed firm-specific templates, better artifact generation, and improved template following.

OpenAI
Built on ChatGPT Enterprise security and controls.
4

### Financial data source controls

The demo shows a Huron Consulting overview request with the Financial data menu open. Separate toggles control source categories, including earnings calls and fund letters; all visible toggles are enabled.

[3:08](https://www.youtube.com/watch?v=opHmX7qClzE&t=188s) [A cropped ChatGPT composer shows part of a company-overview request and an expanded Financial data menu. Visible categories include earnings calls, investor presentations, market prices, analyst estimates, company news, holdings, and fund letters, with blue enabled toggles.](https://assets.ai.engineer/sites/aiecode2025/sha256/afcf706ca9a36982ae2b15340ab39d5b6293d28da2edfcde719de5685d3a8deb.webp)

Financial data
Financials & KPIs
Earnings Calls & Events
Investor Presentations
Market Prices & Trading
Analyst Estimates & Ratings
Company News
Ownership & Holdings
Fund Letters

about Huron Consulting. Give me a company overview covering what I need to know heading into this meeting.

Building the Next Chapter of AI in Finance
Joseph Kim / Head of ChatGPT for Financial Services
OpenAI

Interface text:
+
GPT-6 Astra High
Huron
PDF
Plugins
S&[illegible]
11
5

### Built for asset management

The title card names asset management as a target sector for ChatGPT for Financial Services, highlighting the sector in orange.

[3:58](https://www.youtube.com/watch?v=opHmX7qClzE&t=238s) [White title card with centered text reading Built for in black above asset management in orange. Small presentation navigation controls appear at the lower left.](https://assets.ai.engineer/sites/aiecode2025/sha256/c4d3736c0790fc65d634ac3e170e898a83451b3799cb4ee35917be27da0ba61f.webp)

Built for
asset management

Interface text:
5

### Why we built it

Section 02 introduces why OpenAI built ChatGPT for Financial Services: to close the capability overhang, the gap between improving models and their value in financial professionals’ daily work.

[4:09](https://www.youtube.com/watch?v=opHmX7qClzE&t=249s) [Section 02 divider titled Why we built it, with bold black text inside a large white rounded rectangle over an aqua and mint gradient background.](https://assets.ai.engineer/sites/aiecode2025/sha256/f58da52e7c627c43f6d93ddc0a61a764f260d473a672f57bbd854c93883407f5.webp)

02
Why we built it

6

Interface text:
6

### Model capability is outpacing enterprise value

The 2018–2026 chart depicts model capability outpacing enterprise value; the widening shaded gap is the capability overhang. Kim says closing it requires the right data, tools and an experience that fits how people work.

[4:22](https://www.youtube.com/watch?v=opHmX7qClzE&t=262s) [Chart spanning 2018–2026 with a blue model-capability curve rising increasingly above a black enterprise-value curve. The shaded space between them is labeled Capability overhang. Model milestones annotate the upper curve, while product milestones annotate the lower curve.](https://assets.ai.engineer/sites/aiecode2025/sha256/49665f1281d205804d78f625700ce79d9e86dc92ceb388e14408e39fffa76d22.webp)

Model capability is outpacing enterprise value

Model capability
GPT-1
GPT-2
GPT-3
GPT-3.5
GPT-4
GPT-4o
o1
GPT-4.5
o3
GPT-5
GPT-5.1
GPT-5.2
GPT-5.3
GPT-5.4
GPT-5.5
GPT-5.6
GPT-6 Astra

Enterprise value
API
ChatGPT
ChatGPT Memory
Deep research
Codex Desktop
ChatGPT for FS

Capability overhang

Timeline: 2018, 2019, 2020, 2021, 2022, 2023, 2024, 2025, 2026

7

Interface text:
‹ 7 › ⋮

### Share of work at OpenAI on Codex

Engineering's Codex use grew earlier. In finance, legal and recruiting, Codex's share of the average user's monthly output tokens rose from near zero in early 2026 to 88–91% by June. Kim links the surge to February's desktop app release.

[5:05](https://www.youtube.com/watch?v=opHmX7qClzE&t=305s) [Line chart of the average user's share of monthly output tokens generated on Codex, from August 2025 to June 2026. Labeled endpoints are Engineering 99%, Finance 91%, Recruiting 89% and Legal 88%. Engineering rises earlier, while the other highlighted functions climb sharply during 2026; additional thin gray lines appear in the background.](https://assets.ai.engineer/sites/aiecode2025/sha256/7ad462ab18b1fd458bd55509561e2bc28203af71835c73d07e53c2f5e427a157.webp)

Share of work at
OpenAI on Codex

Legend:
Engineering
Finance
Legal
Recruiting

Y-axis: Average user's share of monthly output tokens generated on Codex
Y-axis ticks: 0%, 25%, 50%, 75%, 100%
Timeline: Aug 2025; Oct 2025; Dec 2025; Feb 2026; Apr 2026; Jun 2026

Endpoint labels:
Engineering (99%)
Finance (91%)
Recruiting (89%)
Legal (88%)

Slide number: 8

Interface text:
‹ 8 › ⋮

### Bundled premium financial data

OpenAI hosts and indexes financial data spanning private companies, U.S. equity prices, filings and earnings calls. Kim says this gives it more control over quality and reliability, while the model still accesses the data through MCP.

[6:00](https://www.youtube.com/watch?v=opHmX7qClzE&t=360s) [Two-column list of bundled financial data: Crunchbase for private-company profiles and funding; PitchBook for private-company data; LSEG for Reuters news; Quartr for earnings-call transcripts and audio, investor presentations and international filings; Fiscal.ai for financial fundamentals and metrics; FMP for U.S. equity pricing and market data through Nasdaq; OpenAI for SEC filings; and Daloopa for reported financial statements and operating metrics. A note states that OpenAI hosts and indexes all data.](https://assets.ai.engineer/sites/aiecode2025/sha256/1c9479f4dcd91a83fd335a1475233566909f55b695fe741c42c107d2f318fd45.webp)

Bundled premium financial data

All data hosted and indexed by OpenAI

Crunchbase
Private company and startup profiles; historical funding data

PitchBook
Essentials private company dataset

LSEG
Reuters news

Quartr
Earnings call transcripts and audio; investor presentations; international filings

Fiscal.ai
Company financials and fundamentals; ratios, operating metrics, and KPIs

FMP
U.S. equity pricing and market data through Nasdaq

OpenAI
SEC filings

Daloopa
Reported financial statements; selected KPIsand operating, segment, and geographic metrics

12

Interface text:
‹ 12 › ⋮

### Bundled premium financial data

OpenAI hosts and indexes the bundle of company data, news, filings and earnings materials. Models access it through MCP; Kim says this setup gives OpenAI more control over reliability and lets it post-train models on the datasets.

[6:33](https://www.youtube.com/watch?v=opHmX7qClzE&t=393s) [Slide listing Crunchbase, PitchBook, LSEG, Quartr, Fiscal.ai, FMP, OpenAI and Daloopa, with coverage spanning private companies, financial statements, market prices, news, filings and earnings materials. A note says all data is hosted and indexed by OpenAI. A broadcast banner appears below the slide.](https://assets.ai.engineer/sites/aiecode2025/sha256/ace8f4bc3e1f9356f540c48e6d247bd2aa727447b716cbd0f76b73d75089b169.webp)

Bundled premium financial data

All data hosted and indexed by OpenAI

Crunchbase
Private company and startup profiles; historical funding data

PitchBook
Essentials private company dataset

LSEG
Reuters news

Quartr
Earnings call transcripts and audio; investor presentations; international filings

Fiscal.ai
Company financials and fundamentals; ratios, operating metrics, and KPIs

FMP
U.S. equity pricing and market data through Nasdaq

OpenAI
SEC filings

Daloopa
Reported financial statements; selected KPIsand operating, segment, and geographic metrics

12

Building the Next Chapter of AI in Finance
Joseph Kim / Head of ChatGPT for Financial Services
OpenAI

Interface text:
12

## Full transcript

[0:12](https://www.youtube.com/watch?v=opHmX7qClzE&t=12s) Hi, everyone. I'm Joseph Kim, and I lead ChatGPT for Financial Services, a new product we launched just one month ago. Many of you have already experienced how tools like Codex have changed the way you build in software engineering. AI has become a much bigger part of the work you do every day. Model intelligence is continuing to improve in-- at an incredible pace, but many people working industries like finance haven't yet experienced that same change in their day-to-day work. I expect that to start

[0:42](https://www.youtube.com/watch?v=opHmX7qClzE&t=42s) changing rapidly over the next few months, and builders like all of you in this room have a big role in making that happen quickly. Taking the intelligence in these models and turning it into products that people can use in their daily work. Today, I will share a little bit about what we learned building ChatGPT for Financial Services and how we're thinking about what comes next. I'll start by showing you what we've built, including a quick demo. Then I'll explain why we built it and why we think now is the right time to build. From

[1:12](https://www.youtube.com/watch?v=opHmX7qClzE&t=72s) there, I'll walk you through how we built it with some technical lessons that I hope you can apply to your own products. Finally, we'll take a look at where models are headed and products are going, and what that could mean for financial work over the next year. Let's start with what we've built.

[1:30](https://www.youtube.com/watch?v=opHmX7qClzE&t=90s) ChatGPT for Financial Services is a tailored version of ChatGPT Work built specifically for financial professionals, starting with investment banking and equity research. It's designed to help people at the large financial institutions get more out of their day-to-day work, similar to how Codex speeds you up in software development. As we built the product with our design partners, Morgan Stanley and Evercore, we focused on three core areas. First, financial data. Financial data is the foundation of financial work, and the work you can

[2:00](https://www.youtube.com/watch?v=opHmX7qClzE&t=120s) do is limited by the quality of your sources. We wanted people to have access to the data they need directly within ChatGPT. Second, an interface built for finance. We wanted the product to feel familiar to financial professionals with ways to inspect source material, explore financial charts, and control the sources behind an analysis. Third, your firm's format. Financial work happens within the context of an organization, and the models and presentations people produce need to follow their

[2:30](https://www.youtube.com/watch?v=opHmX7qClzE&t=150s) firm's format and standards. We made it easier to create those artifacts using firm-specific templates. But they say a picture is worth a thousand words, so I'll stop talking for a bit and show you the product.

[4:11](https://www.youtube.com/watch?v=opHmX7qClzE&t=251s) Wonderful. Now, I'd like to take some time to talk about why we built it.

[4:17](https://www.youtube.com/watch?v=opHmX7qClzE&t=257s) We built ChatGPT for Financial Services to help close what we call the capability overhang. As you can see here, model capabilities have improved at a tremendous rate over the last few years. In just a few years, we've gone from models solving high school math problems to systems capable of solving Millennium Prize math problems. But the value people get from these models in their day-to-day work hasn't kept pace. You read about the breakthroughs in the news, yet for many financial professionals, the way they work hasn't changed nearly as quickly. We call that gap the capability overhang, the

[4:47](https://www.youtube.com/watch?v=opHmX7qClzE&t=287s) distance between what the model is capable of and the value people can actually get from it. Closing that gap takes more than building a capable model. People need the right data and tools, and need-- and they need an experience that fits the way they work. That's where builders like you come in. I want to show you an example of how much of a difference the right product experience can make. This is a chart that shows how Codex usage grew within OpenAI across different, uh, job functions. As you can see, engineers started using Codex early, and that share grew

[5:16](https://www.youtube.com/watch?v=opHmX7qClzE&t=316s) steadily. But look at finance, legal, and recruiting. At the beginning of twenty twenty-six, almost none of their AI output was generated in Codex. By June, that share was around ninety percent. If you look carefully, you'll notice a huge inflection point that happened around February, March of this year. I went back to see what happened around this time, assuming that we must have released a new model, but instead, I found that we had released the Codex desktop app in February. My takeaway is that the product made it easier for people outside of engineering

[5:46](https://www.youtube.com/watch?v=opHmX7qClzE&t=346s) to put these capabilities to work. That's the opportunity we're pursuing in finance, building an experience that helps financial professionals get much more value from the intelligence already available to them. Now let me talk about how we built it.

[6:01](https://www.youtube.com/watch?v=opHmX7qClzE&t=361s) Let's start with data. In finance, the work you can do is limited by the quality of the information you have access to. So we wanted the product to come with the sources professionals rely on, from financial statements and earnings transcripts to market and private company data. We looked at existing MCP connectors for financial data and found that many had issues with latency, errors, or search quality. So we went and licensed a lot of this data directly and built our own hosting and search indices using the same infrastructure as our web index. We still use the

[6:31](https://www.youtube.com/watch?v=opHmX7qClzE&t=391s) MCP protocol to give the model access to the data, but hosting and indexing it ourselves gives us more control over the quality and the reliability of the experience. It also gives us the flexibility to post-train our models to work with these datasets. The analogy I like to use is MCP and skills are like handing an instruction manual to a junior analyst versus post-training, where the model is like having a senior analyst who's been working with the data for over two decades and knows all of its quirks like the back of their hand. MCP and skills give the model

[7:01](https://www.youtube.com/watch?v=opHmX7qClzE&t=421s) access and instructions. Post-training can help it develop more familiarity with the data itself. Once we had that data foundation in place, the next step was making it easy to turn that information into professional-grade work. For financial professionals, that work often takes the form of an Excel model, a PowerPoint presentation, or a written analysis that needs to be reviewed and shared with other people. So we focused on helping the model produce those outputs in a way that fits the firm's existing workflows. We built a template system that gives users access to the firm's

[7:31](https://www.youtube.com/watch?v=opHmX7qClzE&t=451s) approved templates when generating spreadsheets and presentations. The goal is to reduce the work someone has to do to get an output ready for their team or clients. We also added things like interactive financial charts, so users can explore stock prices and financial metrics directly within ChatGPT as they work through an analysis. And we built citation viewers that let people inspect the source passages and tables in-- behind an answer. In finance, you need to be able to check where a number came from and understand the context around it. We

[8:01](https://www.youtube.com/watch?v=opHmX7qClzE&t=481s) wanted that review to be a natural part of the experience. These are the kind of product details that help turn a model's capabilities into work that someone can use. The part I think that is most relevant for people in the room is how we built it. Much of the functionality you just saw was built using the same ChatGPT plugin platform that's available to you as developers. For financial data, we use MCP tools and connectors to give the model access to the information it needs. We use citation metadata and MCP apps to let users inspect their sources, and MCP apps

[8:31](https://www.youtube.com/watch?v=opHmX7qClzE&t=511s) also power interactive financial charts. We use skills together with shared artifact tools to guide w-- financial workflows and produce Excel models and PowerPoint presentations. And for financial onboarding, we leverage the native finan-- uh, plugin onboarding experience. What that means for builders is you c-- is that you can focus on the expertise that makes your product valuable. If you have unique data, an existing tool, or a deep understanding of a particular workflow, you can use these building blocks to bring that into ChatGPT and create an experience around it.

[9:01](https://www.youtube.com/watch?v=opHmX7qClzE&t=541s) Financial services is just one example, but I do think this ex-- this, uh, opportunity extends far beyond finance. I personally believe that we're going to see multiple one-billion-dollar businesses built simply as ChatGPT plugins, and that opportunity will keep evolving as the models become more capable.

[9:18](https://www.youtube.com/watch?v=opHmX7qClzE&t=558s) Last but not least, I'd like to spend the rest of the talk on what comes next and what that means for the products we should be building today.

[9:27](https://www.youtube.com/watch?v=opHmX7qClzE&t=567s) I'd like to start by talking about model capabilities and expectations for the future. Models are improving so quickly that it can be difficult to get an intuitive sense of what comes next. One useful way to build that intuition is to look backwards. Something remarkable about the last couple of years is how predictable parts of that progress have been. Model improvements have come from three broad, sometimes overlapping drivers: scaling compute, improving algorithms, and something we call unhobbling, which is about finding ways to get more useful work

[9:57](https://www.youtube.com/watch?v=opHmX7qClzE&t=597s) out of the models. We'll come back to unhobbling on a lat-- on a later slide. In 2002 and 2003, the most visible driver was pre-training, which meant scaling compute by making the models bigger and adding more data. By leveraging scaling laws, researchers were able to predict model, uh, performance with high accuracy. For example, before GPT-4 finished training, OpenAI researchers were able to predict its final loss as well as its performance on a coding eval. Improvements on the algorithm side have also been surprisingly steady.

[10:27](https://www.youtube.com/watch?v=opHmX7qClzE&t=627s) One study found that the compute needed to reach a given level of mo-- language model performance from scratch fell by roughly half every eight months. Although you can't predict the timing of each individual breakthrough, the cumulative rate of, of improvement follows a surprisingly steady trend. In 2024 and 2025, reasoning became another major driver of progress. We used reinforcement learning and post-training to teach models to work through problems more effectively, and we gave them more compute to think when answering. That opened up another way to improve performance alongside

[10:57](https://www.youtube.com/watch?v=opHmX7qClzE&t=657s) pre-training. However, for folks that don't know, at OpenAI, we ran into problems with pre-training and therefore had difficulty scaling the model. In 2026, we've been able to solve many of these issues, and we've successfully trained larger and more capable models, as you've seen through GPT-5.6 Sol and GPT-6 Astra. Throughout these years, researchers have raised concerns about potential barriers to further scaling. For example, in 2024, many researchers were concerned about a potential data wall where we would run out of human-generated

[11:27](https://www.youtube.com/watch?v=opHmX7qClzE&t=687s) text to train on. But techniques like synthetic data generation, learning from verifiable rewards, as well as algorithmic improvements to learn more from each individual token have essentially solved this problem. We've also solved concerns around things like context window length using things like sparse attention and compaction. Looking ahead to 2027, I strongly believe that the trend lines of the past will continue, and that we'll train larger, more efficient, and more capable models. We clearly still have very difficult problems to solve, especially

[11:57](https://www.youtube.com/watch?v=opHmX7qClzE&t=717s) around things like alignment and safety, but like we have in the past, I believe that human ingenuity, increasingly assisted by capable models, will help us make progress on those problems too. So how good will models be at the end of twenty twenty-seven? Let's use math as an example. In two thousand twenty-three, GPT-4 scored seven hundred out of eight hundred on SAT high school math. By two thousand and five, an OpenAI model reached gold medal status on the International Math Ma-Mathematical Olympiad. And in twenty-- in early twenty twenty-six, an

[12:27](https://www.youtube.com/watch?v=opHmX7qClzE&t=747s) internal model disproved a long-standing conjecture related to a problem that mathematicians have been working on since nineteen forty-six. More recently, an internal model proposed a solution to the Navier–Stokes Millennium Prize Puzzle, um, Problem, and we also saw-- shared seven hundred and nineteen mathematical research manuscripts across three hundred and seventy-two families of results. In just a few years, we've gone from a strong score on a high school math exam to AI systems producing original research at the frontier of mathematics. If we see another year of progress roughly in

[12:57](https://www.youtube.com/watch?v=opHmX7qClzE&t=777s) line with what we've been seeing over the past couple years, I expect that by end of twenty twenty-seven, models will outperform even leading human experts on a growing range of difficult tasks across different fields and specialties. Mathematics gives us an early glimpse of what that could look like.

[13:13](https://www.youtube.com/watch?v=opHmX7qClzE&t=793s) And while the capability curve may look gradual, the change in how we work can feel sudden. I call this the threshold phenomenon. For any given task, there's a threshold of usefulness, the point where a model becomes capable and reliable enough to save you time. That threshold depends on the work you're doing and how much checking or human review it requires. Below that threshold, improvements can be hard to appreciate. The model might produce better answers, make fewer mistakes, or get further through a task. But if checking and fixing its work still takes as long as doing it yourself, your practical assessment hasn't changed.

[13:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=823s) It's still not useful. Meaningful improvements in capability can feel like s-- like little progress. Then the model crosses that threshold. Suddenly it produces something you can use. The time you spend reviewing it becomes much smaller than the time it saves you. You start using it regularly, and eventually you become comfortable delegating that task. That transition can feel like an enormous leap, even if the underlying improvement was similar to the previous few improvements. What changes the progress became useful to you? This helps explain why model progress sometimes feels sudden, even when parts

[14:13](https://www.youtube.com/watch?v=opHmX7qClzE&t=853s) of the underlying trend have been steady and predictable. And because difficult tasks have different thresholds, we should expect those moments to arrive at different times across the work that we do. So as you think about the products you're building, I encourage you to ask yourself, "What model capabilities am I underappreciating today because they aren't quite useful yet? Where might those capabilities be a year from now?" Then think about what you would build if those capabilities crossed the threshold of usefulness and became reliable enough to delegate. How would the interface change? What

[14:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=883s) would the user still need to control, review, or approve? Those are product design questions we can start working on now. We want to have the right experiences ready so that when the models cross that threshold, people can take advantage of the capabilities right away, and so you can stay one step ahead of your competitors. So what does this mean for finance? I think there are two forces that will push more financial work across that threshold of usefulness. I think of them as the rising tide and the waves on top. The rising tide is the underlying improvements, uh, uh, to the underlying model.

[15:13](https://www.youtube.com/watch?v=opHmX7qClzE&t=913s) Historically, we haven't dedicated a significant amount of training time or compute specifically to finance. Much of the financial capabilities you see today have come from building larger, more capable models and from transferable skills developed in areas like mathematics, coding, and reasoning. As these general capabilities improved, they lift the performance across many kinds of financial work. The waves are what happen when we focus directly on finance. My team is leading an effort to invest significantly more in finance-specific post-training. The idy-- idea here is

[15:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=943s) to build on those broad capabilities and train the model more directly on the work people do in finance. That can create peaks in performance on particular tasks on top of the rising tide. Over the next three months, we're focused on improving fin-- general financial reasoning and the quality of the work these models produce, analyzing financial information, building Excel models, and creating PowerPoint presentations. These are the kinds of tasks that make up a significant percentage of the daily work of a junior analyst, and we want the output to be accurate, well-structured, and increasingly useful in real workflows.

[16:13](https://www.youtube.com/watch?v=opHmX7qClzE&t=973s) Beyond that, we plan to invest more in the kinds of judgment that senior bankers, analysts, and investors bring to more difficult questions. This includes taking in large amounts of information, deciding what matters, and essentially trying to predict the future, forecasting a company's earnings, assessing where commodity prices might go, or evaluating which businesses are attractive investments. Our goal is for models to match the be-- the performance of the best human experts on these types of financial tasks sometime in twenty twenty-seven. We expect that progress to come from both forces working together,

[16:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=1003s) the underlying model continuing to improve and our team training it more directly for the work people do in finance. We've talked about how models are becoming more capable and how we can train them to perform better on specific tasks, but I wanna come back to one third driver I mentioned earlier, unhobbling. We've already seen a few examples. ChatGPT, um, used RLHF to teach models to follow instructions and have conversations with us. Tools allow them to search for information, write and run code, and interact with the software we use. Each of these makes the model more useful in the real

[17:13](https://www.youtube.com/watch?v=opHmX7qClzE&t=1033s) world. I think of the next form of unhobbling will be proactive and persistent agents. Today, a lot of our interactions with AI still happen when we open a conversation and tell it what to do next. A persistent agent can hold onto a goal, remember the relevant context, and continue working in the background. Once you've given it a direction, it can make progress even when you aren't actively prompting it. The other reason persistence matters is it changes the API or the way people work with the agent. Today, getting the most out of AI agents requires learning a new workflow, learning how to prompt it,

[17:43](https://www.youtube.com/watch?v=opHmX7qClzE&t=1063s) and learning how to use the tools. A persistent agent can work in a way that's more familiar. You can give it a project, share context, and ask for an update. It's much more similar to how humans work with other humans, and therefore makes the AI more accessible to people, um, that don't develop those initial skills. Let me close with three bets for finance in twenty twenty-seven. These are just a couple of my expectations for what comes as the models improve. First, people will rely rarely, if ever, on Excel and PowerPoint to make financial models or pitch decks.

[18:14](https://www.youtube.com/watch?v=opHmX7qClzE&t=1094s) Second, AI financial analysts will start to feel like human coworkers. They'll remember your preferences and context from the previous days, take ownership of assignments, proactively follow up, and keep working when you're focused on something else. Third, models will start to outperform the best human experts on critical financial tasks. That includes things like forecasting earnings and developing investment ideas. I expect models to become better than the best people at making these economically important forward-looking judgments. Those are my three bets for twenty twenty-seven, and if this is where we expect to be in twenty twenty-seven, I'll

[18:44](https://www.youtube.com/watch?v=opHmX7qClzE&t=1124s) leave you with one last question. What would-- what should you start building today to be ready for inevitable thresholds that the models will cross tomorrow? Thank you.
