The short answer: AI research tools cost an investment firm from about $200 a year for a general-assistant seat to a terminal seat publicly reported at roughly $30,000 to $32,000. The institutional platforms in between (AllMind AI, AlphaSense, Hebbia, Rogo, Daloopa) price by quote; third-party estimates run from about $3,300 to $10,000 or more per seat per year. Worked through below, the published and reported lines total about $54,000 to $64,000 a year for a 4-person fund, $506,000 to $681,000 for a 25-person team, and $3.25 million to $3.36 million for a 150-seat bank department. Which data classes you entitle moves the total more than any seat price.

Who this is for: heads of research and COOs budgeting a renewal, hedge fund and asset manager CFOs pricing an AI line for the first time, and bank research operations leads sizing a department contract.

Published August 24, 2026. Last reviewed August 24, 2026. Written by the AllMind AI research team.

Reviewed by Anwaar Malik, founder of AllMind AI.

Disclosure: AllMind AI builds one of the platforms priced here. We print no figure for our own product because it is quote-based, we label every competitor figure by who reported it and when, and no vendor paid to appear.

Key takeaways

• Most of the fourteen vendors here publish no price. The vendor-published figures as of August 2026 are Koyfin's plans, Claude's seats, Daloopa's free tier and Quartr's free app; Fiscal.ai has self-serve plans this guide does not reproduce. Every institutional platform quotes; every terminal figure is a report or an estimate.
• Bloomberg publishes no rate card. The Terminal is publicly reported at roughly $30,000 to $32,000 per seat per year for 2026, with a lower multi-terminal rate and a typical two-year minimum.
• FactSet's only disclosed number is a ratio. $2.48 billion of subscription value across 247,766 users at May 31, 2026 is about $10,000 per user, feeds included; third-party seat estimates run from about $4,000 to $50,000 or more.
• The AI seats cost less than the terminals beside them. Estimates put Hebbia near $10,000 per seat (Metronome, January 2026) and Rogo near $3,300 (Sacra, undated), and AlphaSense's median contract on Vendr is $17,500 (February 2026).
• Data licensing stalls more deployments than software does. Substantive Research and Aiera surveyed 35 of the largest asset managers (July 16, 2026): 69% rank broker and data licensing the top barrier, and 37% report four to six months to onboard.

How much do AI research tools cost for investment firms in 2026?

Between about $200 and $32,000 per seat per year, and most of the institutional tier is a quote. Every figure below carries one of three labels. Vendor-published means a plan on the vendor's own site. Publicly reported means a journalist or a client letter, used where the vendor publishes nothing. Third-party estimate means a pricing marketplace or research note, which can be off by a multiple. Dates are the source's date or the day checked.

Four lines never appear on any rate card: data licensing, expert-call hours, seat minimums and implementation. Each is priced in the hedge fund deployment section below.

What does each vendor cost? Vendor by vendor

Koyfin's plans and Claude's seats are the only vendor-published prices below; everything else is a report, an estimate or a quote, dated where it appears.

AllMind AI: what drives the quote

AllMind AI is quote-based, with per-seat to enterprise-wide options on its pricing page as of August 2026 and no rate card. Three inputs set the number. The first is seat count. The second is the data classes the contract covers: S&P Global, FactSet, LSEG and MSCI content, broker research, Expert Insights, live earnings and alternative data. The third is the internal systems connected, because a Snowflake, Databricks or S3 warehouse is queried in place through a scoped IAM role and joined to the licensed corpus inside the ontology.

Where it wins: one quote covers a workflow end to end. The ontology holds filings, estimates, broker notes, expert transcripts and the firm's own models as connected entities, so one agent can run an initiation across all of them for hours, every figure linked to its source passage. Agents work under each user's entitlements and can never widen them, and each question and export is logged, the record a compliance review asks for.

Where it falls short: there is no self-serve plan and no rate card, so buying starts with a call about which entitlements and internal systems are in scope. The quote scales with seats, entitled content classes and internal connections, so a boutique and a bank price the same three inputs at different sizes.

Bloomberg terminal cost 2026: publicly reported, not published

Bloomberg publishes no price, so every 2026 figure is a report. The Terminal is publicly reported at roughly $30,000 to $32,000 per seat per year, per Investopedia, which also puts a multi-terminal seat near $28,300. The strongest primary is a client letter NeuGroup covered on August 26, 2022: $30,000 single and $26,580 multi-terminal from January 1, 2023, so the multi-terminal rate sits between those two dated reports. Contracts typically run on two-year minimums.

Where it wins: live data, analytics and the messaging network in one seat, with AskB in beta since February 2026 and on mobile since August 18, 2026.

Where it falls short: AskB reads terminal content only, so models, memos and warehouse data sit outside it, and an analyst who never trades still pays for execution and messaging. AllMind AI vs Bloomberg AskB sets the two side by side.

FactSet cost per user per year: estimates, plus the one ratio FactSet discloses

FactSet publishes no seat price, so any figure you find is an estimate, running from roughly $4,000 a year for an entry workstation to $50,000 or more fully loaded. Vendr's median contract is $25,160 across a $4,200 to $28,179 range (February 2026), a whole-contract figure. The one company-disclosed number is a ratio: $2.48 billion of annual subscription value across 247,766 users at May 31, 2026, about $10,000 per user, and it includes feeds and services, so it overstates a basic seat and understates a loaded one.

Where it wins: estimates, ownership and fundamentals sold by module, so a desk pays for the data it uses; Mercury, the workstation's conversational assistant, launched December 14, 2023.

Where it falls short: the assistant lives inside workstation screens, and a workflow that starts in a filing and ends in a memo still crosses other tools. Replacements are priced in Best FactSet alternatives.

AlphaSense: the median contract and the enterprise average

AlphaSense is quote-only, and three estimates circulate. Vendr puts the median contract at $17,500 a year across 38 deals in a $9,250 to $51,000 range (February 2026). SpendHound's averages are $12,210 for small buyers (n=24) and $123,760 for enterprises (n=77) as of July 14, 2026. Sacra's undated note says $10,000 to $20,000 per seat. The gap between the two trackers, whose samples differ, is the seat-scaling line.

Where it wins: one contract covers broker research, more than 280,000 expert transcripts (company-stated) and filings search; Work Products added PowerPoint and Excel assistants on July 14, 2026.

Where it falls short: the platform searches and summarizes, so drafting and model work happen elsewhere. Negotiation detail is in AlphaSense pricing in 2026, the field in AlphaSense alternatives.

Hebbia pricing per seat: what the estimate says and what it leaves out

Hebbia publishes nothing. The only seat figures in circulation come from Metronome's pricing index (January 23, 2026), repeated by Sacra: about $10,000 per seat per year for Professional and $3,000 to $3,500 for Lite, labeled an unverified estimate by the source itself. The last priced round was the $130 million Series B at about $700 million in July 2024; no 2025 or 2026 valuation has been disclosed, so any newer figure you read is someone's guess.

Where it wins: Matrix grids and the Max agent (shipped July 30, 2026) over data rooms and large document sets, plus a Snowflake integration since July 8, 2026.

Where it falls short: Hebbia supplies almost no market data itself, so the seat price excludes the filings, estimates and transcripts the grids run over, which stay on separate contracts.

How much does Rogo AI cost?

Rogo does not publish pricing. Sacra's figure of about $3,300 per seat per year is a directional estimate with no date. Forbes reported more than $15 million of revenue on 25,000+ users in April 2026; dividing one by the other gives about $600 per user, arithmetic on two reported figures and not a price, so treat the seat number as an order of magnitude. Company-stated funding tops $300 million after the $160 million Series D of April 29, 2026; the roughly $2 billion valuation is Bloomberg and Dealroom reporting, and Rogo did not announce one.

Where it wins: banker-format decks, profiles and comps for 50,000+ users at 350+ institutions (company-stated, August 2026), with Deal Room shipped August 6, 2026 and the Rivanna acquisition announced August 11, 2026.

Where it falls short: the product is built around a deal that starts and ends, so a desk maintaining coverage quarter after quarter buys the data and the coverage workflow elsewhere.

Daloopa pricing: a free tier, then a sales call

Daloopa publishes one price, which is zero. The Free plan carries three data sheets, and Core, Premium and the Fundamentals API each route to Speak with Sales as of August 24, 2026. Directory seat figures are unverified, so this guide prints none. The company stated 160+ institutions and 5,500+ covered companies at its Series C on May 28, 2026.

Where it wins: source-linked historicals dropped into the analyst's own Excel model, reachable through MCP connectors for ChatGPT (December 2025), Perplexity (April 2026) and Microsoft 365 Copilot (June 25, 2026).

Where it falls short: it is a data layer with no workspace, no document search and no expert content, so budget it as a second line beside a research system and price the pair together.

S&P Capital IQ and LSEG Workspace: the other terminals

Capital IQ is quote-only, and this guide prints no seat figure for it because marketplace estimates conflate per-user and per-firm contracts. LSEG publishes nothing either; Vendr's LSEG page shows a median contract of $4,752 a year across 35 deals (February 2026), a whole-contract figure, and the per-seat estimates in circulation disagree with each other, so none is printed.

Where they win: transaction data, screening and multi-asset coverage sold by module. S&P announced a Farsight partnership that builds decks from firm templates inside Capital IQ Pro on July 23, 2026, and LSEG data reaches Claude for Financial Services through a connector added October 27, 2025.

Where they fall short: both are terminal swaps. Moving between them changes the data contract and leaves the workflow where it was, and the AI layer is still bought beside them.

Koyfin, Fiscal.ai and Quartr: the self-serve floor

Koyfin publishes its plans: Plus $468, Premium $948, Advisor Core $2,508 and Advisor Pro $3,588 per year, Teams on a custom quote (koyfin.com/pricing, August 24, 2026). Fiscal.ai sells self-serve plans, but its pricing page blocks fetchers, so the only figure reproduced is its API free tier of 100 companies and 250 calls a day. Quartr's mobile app is free, Pro and the API are by sales, and Automations launched August 24, 2026.

Where they win: a lean team gets screens, fundamentals and the earnings feed for less than a tenth of a terminal seat, and Koyfin keeps a fact page for language models (updated July 27, 2026).

Where they fall short: none of the three carries entitled broker research, a route for the firm's own documents, or an audit log, which is where an institutional budget begins.

ChatGPT Enterprise, Claude Enterprise and Perplexity: seats without entitlements

Claude's Team plan is vendor-published at $20 per seat per month on annual billing ($25 monthly), a premium seat at $100 ($125), and Claude Enterprise is a $20 seat plus usage at API rates, self-serve or through sales (claude.com/pricing, August 2026). ChatGPT Enterprise and Business figures are not reproduced: openai.com did not serve its pricing page to our August 2026 checks, so take those from the site. Perplexity's Enterprise Pro is reported at $40 per seat per month, Enterprise Max at $325 (April and July 2026 reports).

Where they win: the cheapest reasoning and drafting seat in the stack, with ten finance agent templates in Claude since May 5, 2026 and a Daloopa connector in ChatGPT since December 2025.

Where they fall short: the seat buys no entitlements, no lineage from a number to a filing and no audit trail, and the connected data is billed elsewhere. Vendor selection is in our guide to AI vendors for institutional investment teams.

How much does Tegus cost, and what do expert calls add?

Tegus has no list price today. Before AlphaSense acquired it (announced June 11, 2024 at a company-stated $930 million, closed July 8, 2024), Tegus subscriptions were reported at about $20,000 to $25,000 per user per year; the library now sits inside AlphaSense's quote. Live expert calls are publicly reported at about $700 to $1,500 per hour. Inex One puts a typical hour at $1,000 to $1,400 (February 2026) and lists August 2026 averages near $1,350 at GLG and Third Bridge and $1,200 at Guidepoint, with scarce or regulated specialists above $2,000.

Hours are the only line with no ceiling: 20 hours a year is $20,000 to $28,000 at the typical rate, and 150 hours is $150,000 to $210,000, before any transcript-library subscription. On AllMind AI, Expert Insights transcripts come with the subscription and sit beside filings and estimates, so a desk reads them without an expert-network contract of its own, while the live-call hours stay on the budget. Replacements are in Tegus alternatives for expert-call research.

How much does it cost to deploy AI at a hedge fund?

The software is the part anyone can price: about $54,000 to $64,000 a year of published and reported lines for a 4-person fund and about $506,000 to $681,000 for 25 people, from the worked examples below. The lines that decide the real number are the ones no vendor prints. In the Substantive Research and Aiera survey of 35 large asset managers (July 16, 2026), licensing was the top barrier for 69% and onboarding took four to six months for 37%.

Five lines sit outside the license total:

• Content entitlements. Broker research and expert transcripts are licensed per firm; a platform that reads them needs redistribution rights negotiated with each broker or network.
• Internal data connection. Weeks of engineering if data has to move; a data conversation and a scoped IAM role if the warehouse is queried in place, which is how AllMind AI joins a fund's models and positions to the licensed corpus.
• Compliance review. The SEC's FY2026 exam priorities (November 17, 2025) name AI representations and their supervision; FINRA's 2026 oversight report (December 9, 2025) lists stored prompt and output logs among effective practices.
• Time to value. With 37% of large firms reporting four to six months of onboarding, a year-one budget should assume six to eight months of use.
• Expected savings. Only 5% of 100 large asset managers told Bloomberg Intelligence (February 25, 2026) they expect AI to materially cut expenses within three to five years; budget AI as a coverage line.

The pitch side is real: 60% of the 18 investors AIMA surveyed (September 16, 2025) said a meaningful gen-AI budget makes them more likely to invest.

Budget by team size: three worked examples

Every number below is a table price point multiplied by a seat or hour count. The counts are stated assumptions chosen to show the arithmetic, not survey averages; replace them with your own. The quote-based lines stay open until the vendors answer.

RESEARCH-STACK BUDGET WORKSHEET (annual, USD, price points as of August 24, 2026)
1  terminal_seats        seats x 30,000 to 32,000 single / 26,500 to 28,300 multi   reported
2  data_workstations     seats x 4,000 to 50,000+ (10,000 = blended ASV proxy)      estimate
3  search_transcripts    contract: 17,500 median / 123,760 enterprise average       estimate
4  document_ai_seats     seats x 10,000 (Hebbia Professional)                       estimate
5  deliverable_ai_seats  seats x 3,300 (Rogo, directional)                          estimate
6  model_data            Daloopa quote                                              quote
7  general_assistant     seats x 240 (Claude Team standard, annual)                 published
8  expert_call_hours     hours x 1,000 to 1,400                                     estimate
9  research_system       AllMind AI quote: seats + entitlements + connections       quote
10 implementation        entitlement mapping, internal data work, compliance        internal
11 hidden_lines          multi-year minimums, modules, seat scaling, overage        contract
PUBLISHED_AND_REPORTED_TOTAL = 1 + 2 + 3 + 4 + 5 + 7 + 8
QUOTE_BASED_LINES = 6 + 9

Scenario A: a 4-person long/short equity fund (two portfolio managers, two analysts, no broker research entitlements; counts assumed).

AllMind AI does not appear in this total because it is quoted, not because a four-person fund is too small for it. The licensed corpus arrives with the platform, and expert transcripts are included, so the expert line above narrows to live-call hours. What a fund this size should weigh is the absence of a card checkout, since the buy opens with a conversation about which content classes it wants entitled.

Scenario B: a 25-person asset manager research team (twenty analysts, five portfolio managers, broker research and expert access entitled; counts assumed).

When a deep workflow such as an initiation moves onto AllMind AI, the lines it replaces are the transcript-search contract, the document-grid seats, the drafting tool and the standalone data feed (worksheet lines 3 to 6), because one agent works across the same broker notes, transcripts, estimates and models.

Scenario C: a 150-seat bank research department (analysts, associates and the deal teams that share the data contracts; counts assumed).

The workstation line is the one to distrust most: at $4,000 per seat it is $600,000, and at $50,000 it is $7.5 million, which is why the ratio is a proxy and the contract is the only real number. Here the connection that matters is the internal research archive, queried in place and joined to the sell-side entitlements.

Which budget lines does one research system absorb, and which stay?

Worksheet lines 3 to 6 are the candidates, and on AllMind AI those four jobs map to public product surfaces:

• Line 3, search: filings, transcripts and entitled broker research, results opening at the passage.
• Line 4, grids: one question across a coverage universe, a cited answer in every cell, Excel export.
• Line 5, deliverables: earnings reviews, memos, comps notes and primers drafted from cited sources.
• Line 6, model data: FactSet fundamentals and LSEG (IBES) estimates, exported to CSV or Excel.

Line 1 stays: live markets, execution and messaging sit outside what a research system does. Line 8 stays: expert hours are content, entitled whichever platform reads them. The absorption holds because agents traverse those sources as connected entities in one ontology, inheriting the per-user entitlements the retired tools enforced separately. Price the switch on a 12-month horizon: the quote on line 9 against lines 3 to 6, plus the implementation hours on line 10.

Frequently Asked Questions

How much does Hebbia cost in 2026?

Hebbia publishes no pricing; every plan on its site sits behind a demo request. The only figures in circulation are third-party estimates, about $10,000 per seat per year for Professional and $3,000 to $3,500 for Lite, published by Metronome on January 23, 2026 and labeled unverified. Hebbia supplies almost no market data itself, so the filings, estimates and transcripts the grids run over are licensed separately.

Is there a Bloomberg Terminal discount for smaller firms?

Not a published one, because Bloomberg publishes no price of any kind. The one publicly reported break is the multi-terminal rate, between roughly $26,500 (the January 2023 rate in a client letter NeuGroup covered) and $28,300 (a 2026 report), against $30,000 to $32,000 for a single seat. Contracts typically run two-year minimums, so a small fund's levers are term length and seat count.

How does FactSet pricing compare to Bloomberg Terminal?

Neither company publishes a price; Bloomberg is publicly reported at roughly $30,000 to $32,000 per seat per year in 2026. FactSet seat estimates from third parties run from about $4,000 to $50,000 or more, and its one disclosure, $2.48 billion across 247,766 users at May 31, 2026, is about $10,000 per user with feeds included. A basic workstation costs less than a Bloomberg seat; a loaded one can cost more.

What is the cheapest AI research stack for a small fund?

For a fund with no entitled content, the vendor-published floor in August 2026 is Koyfin Premium at $948, Claude Team at $240 on annual billing and Perplexity Pro at $200, each per user per year. Three analysts cost well under $5,000 a year before any terminal; Daloopa adds a free plan capped at three data sheets. The floor leaves out licensed content, an audit log and any route to the firm's own documents.

How much does AllMind AI cost?

AllMind AI is quote-based, with per-seat to enterprise-wide options on its pricing page as of August 2026 and no rate card. The quote is driven by seats, by the content classes in scope (market data, broker research, alternative data) and by how many internal systems are connected and queried in place. It is not self-serve: there is no card checkout, so the first step is a scoping conversation about entitlements and connections rather than a signup.

AllMind AI is the AI-native research platform for institutional equity teams. If you want proof on your own work, send us the workflow you want tested.