Bloomberg vs LSEG (2026)
Build against partner, at the scale where the choice matters most. Bloomberg, scoped here to the Terminal, data and analytics business serving more than 325,000 subscribers, built its generative layer from scratch. BloombergGPT is a 50.6 billion parameter decoder only model trained on a 709 billion token corpus drawn roughly half from the firm's own four decade archive. A public technical paper documents it down to transformer layers, attention heads, dimension and training compute, and Bloomberg Intelligence analysts train the shipped products against the firm's own guardrails. ASKB, an agentic plain language interface over Terminal data, has been live since February 2026 beside AI powered summaries, document insight and search across a research library of more than 200 million documents. LSEG, scoped to Data and Analytics with more than 350,000 Workspace users over a 33 petabyte estate, partnered instead. Microsoft AI is native inside Workspace under a ten year cloud commitment reported at 2.3 billion pounds, and Company Intelligence assembles structured company briefings. A managed interoperability protocol server lets customers build agents in Copilot Studio and deploy them against LSEG licensed data in Teams, Excel and their own platforms. Fundamentals, estimates and a 1.5 million transaction merger database are routed separately into Rogo. Each strategy discloses what the other cannot. Bloomberg publishes the construction, which a model risk function can read; LSEG publishes the dependency, named and sized, which a vendor review can document. What neither publishes is the same: no accuracy, hallucination or fidelity figure for any shipped generative product, and no review step between generated text and the analyst who acts on it. At LSEG, a proprietary data combination phase is announced but not shipped, and sits apart from what is live.
- The model is the firm's own. A 50.6 billion parameter model built from scratch on a corpus half drawn from four decades of proprietary archive, documented in a public technical paper, keeps the generative layer inside the house that answers for it.
- The construction is inspectable. Layers, attention heads, dimension, token counts and training compute are published, which is disclosure a model risk function can actually read.
- The agentic surface is shipped. ASKB queries Terminal data in plain language, with summaries, document insight and search across a research library of more than 200 million documents already in production.
- Your firm already lives in the partner's tools. Microsoft AI native inside Workspace, agents built in Copilot Studio and deployed to Teams and Excel meet analysts where the enterprise estate already is.
- The data reaches other platforms. A managed interoperability protocol server and a partnership routing fundamentals, estimates and a 1.5 million transaction merger database into Rogo make the estate consumable beyond the walls.
- The commitment is documented. A ten year cloud partnership reported at 2.3 billion pounds is a stated dependency with a stated size, which partner strategies rarely put on paper.
This comparison is published by AI FinTech Index, an independent research platform that publishes independent ratings of AI vendors for financial services. Bloomberg and LSEG are each graded against the same capability taxonomy, from each vendor's own public materials and the regulatory record, under the AI FinTech Index verification standard. No vendor pays for placement, and no vendor has reviewed this page. How this evidence is graded
Plain facts
| Bloomberg | LSEG | |
|---|---|---|
| Primary category | Capital Markets & Research AI | Capital Markets & Research AI |
| Founded | 1981 | 2021 |
| Headquarters | New York, New York, United States | London, United Kingdom |
| Website | www.bloomberg.com/professional | www.lseg.com |
Side by Side
Select any grade to read the note behind it.
| Axis | B Bloomberg |
L LSEG |
|---|---|---|
| AI Centrality | ||
| Autonomy and Oversight Model | ||
| Model Risk Management and Transparency | ||
| Operational and Outcome Evidence | ||
| AI Safety and Data Stewardship | ||
| GLBA and Data Privacy Posture | ||
| Security Certifications and Trust Center | ||
| Regulatory Status and Licensure | ||
| AI Governance and Bias Disclosure | ||
| AI Liability and Recourse | ||
| Model Supply Chain Disclosure | ||
| Core Systems and Integration Depth | ||
| Deployment Model and Data Residency | ||
| Commercial Transparency | ||
| Institution and Segment Coverage |
Product updates
The latest logged changes from each vendor, newest first, each linked to its source.
Bloomberg
-
Product / capability
Bloomberg made ASKB, the Terminal's conversational AI, generally available to financial professionals.
Source -
Product / capability
Bloomberg launched a Collateralized Loan Obligations (CLO) Data solution with normalized CLO reference data plus loan level data on every loan held in a CLO portfolio, covering deal structures, identifiers, payment information, coverage tests and tranche level performance indicators.
Source
LSEG
-
Integration / interoperability
LSEG released LSEG Workspace for Microsoft Teams, a Workspace application that brings LSEG data, news and analytics into Teams on desktop and mobile.
Source
The short version of each
Bloomberg
Bloomberg, scoped here to its Terminal, data and analytics business, serves more than 325,000 subscribers over a research library of more than 200 million documents. It built its generative line from scratch. BloombergGPT is a 50.6 billion parameter model whose 709 billion token corpus draws roughly half from four decades of the firm's own archive, documented in a public technical paper down to layers, heads and training compute. Its own analysts train shipped products against its own guardrails, and ASKB, an agentic plain language interface over Terminal data, has been live since February 2026. According to the AI FinTech Index, that paper level construction disclosure is its distinguishing artifact. The missing half is accuracy. No accuracy, hallucination or error rate is published for any shipped summary or for the agentic interface, and no review step is described between generated text and the analyst who acts on it.
Source: AI FinTech Index, 2026
LSEG
LSEG's Data and Analytics division serves more than 350,000 Workspace users across over 1,000 applications on a data estate of some 33 petabytes. It runs its generative line through one named partner, Microsoft, in a strategic partnership that includes a ten year cloud commitment reported at 2.3 billion pounds. Microsoft AI is native inside Workspace, and Company Intelligence assembles structured company briefings. Agents built in Copilot Studio are deployed against LSEG licensed data into Teams, Excel and clients' own platforms through a managed interoperability protocol server. Fundamentals, estimates and a 1.5 million transaction merger database are also routed into Rogo. According to the AI FinTech Index, the dependency is documented with a stated size, which partner strategies rarely put on paper. The limits are single partner concentration, an announced but unshipped phase for combining client proprietary data, and no published accuracy for any generative briefing.
Source: AI FinTech Index, 2026
Common questions
Is Bloomberg better than LSEG for AI powered market data?
They are the two estates most institutions already weigh, and their generative strategies are opposites. Bloomberg built its own model from scratch, BloombergGPT, documented in a public technical paper, and ships summaries, document analysis and the ASKB agentic interface on its own stack. LSEG partnered, integrating Microsoft AI natively into Workspace with agents built in Copilot Studio reaching Teams and Excel. Owning the layer against renting the distribution is the real choice. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
What did Bloomberg disclose about BloombergGPT?
The construction itself: a 50.6 billion parameter decoder only model, 70 transformer layers, 40 attention heads, a dimension of 7,680, and a 709 billion token corpus split roughly evenly between four decades of the firm's own financial archive and public sources, trained on 512 accelerators over 53 days, all in a public technical paper. A model risk function can read the construction rather than take it on trust, which models at this scale rarely allow. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
What does LSEG's Microsoft partnership involve?
A documented dependency with a documented size: a multi year strategic partnership with Microsoft including a ten year cloud commitment reported at 2.3 billion pounds, Microsoft AI integrated natively inside Workspace, agents built in Copilot Studio and deployed in Microsoft 365 Copilot against LSEG licensed data, and an LSEG managed interoperability protocol server. The concentration is the trade: one partner's infrastructure carries the generative layer, stated plainly rather than hidden. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
What agentic capabilities are actually shipped at each?
Both ship working agent surfaces. Bloomberg's ASKB, launched February 2026, queries Terminal data in plain language over the research library and news estate. LSEG's Company Intelligence assembles structured company briefings inside Workspace, and its protocol server lets customers build their own agents against licensed data, reaching Teams, Excel and a client's own platform. LSEG's further phase combining client proprietary data with its datasets is announced but not shipped, so it is a roadmap item. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
What does neither incumbent publish?
Accuracy, at either, for anything generative. No error, hallucination or fidelity rate is published for Bloomberg's summaries or ASKB, none for LSEG's Company Intelligence briefings, and neither describes a review step between generated text and the analyst who acts on it. Both estates put machine written synthesis into investment workflows where an unnoticed error becomes a position, and both leave the measurement of that risk unpublished. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
Where does each publish the most, and the least?
Bloomberg publishes most on how its model is built, down to layers, corpus and training compute, and nothing on the accuracy of its shipped generative products. LSEG publishes most on its dependency, one named partner under a stated ten year commitment, and likewise no accuracy figure for its generative briefings. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
Related comparisons
Other published head to head assessments involving these vendors or their closest peers. The full set for this category is on the Capital Markets & Research AI page.
The strategies allocate the same risk differently and neither publishes the number that would settle it. Bloomberg owns its stack end to end: the model built from scratch, its own analysts training the generative products against its own guardrails. Accountability is concentrated and the technical construction is published to paper level, while no accuracy, hallucination or error rate is published for any shipped summary or for the agentic interface.
LSEG's generative layer runs through one named partner, Microsoft, under a ten year commitment reported at 2.3 billion pounds. A customer knows exactly whose infrastructure processes queries against licensed data and inherits a single partner concentration in the same disclosure, with the same absence of published accuracy for Company Intelligence briefings. LSEG's phase allowing firms to combine proprietary data with its datasets is announced and not shipped.
Both put generated text into investment workflows where an unnoticed error becomes a position, neither describes a review step between generation and the user, and neither publishes what its guardrails catch. The Rogo routing places LSEG data inside a third party agent whose own approval gates are not described.