Paper Instruments

The best AI tools for accountants, compared by the job

This guide compares the eight tools accountants weigh: Numeric, FloQast, BlackLine, Basis, Copilot in Excel, Claude for Excel, ChatGPT Work and Ultramarine Work. It reads each on what it does to a month-end close, which models run it, what it costs, and when to pick it over another.

Checked on

At a glance

ProductBest forAvailable onStarts atFree access
NumericClose management, recs and flux on ERP dataCloud platform with ERP integrationsNot published, quoted by volume and seatsNone listed on the pricing page
FloQastClose automation with agents and SOX controlsCloud platform with ERP integrationsQuote, priced on outcomes, not per userNone mentioned
BasisAgents preparing firm client work for reviewAgent platform linked to firm systemsNot publishedNone mentioned
Ultramarine WorkDrafting and checking close files you reviewWeb app and desktop app for Mac and Windows$5 a month (Starter)7 days free on Starter and Plus
BlackLineEnterprise reconciliations with agent preparationEnterprise cloud platformNot publishedNone mentioned
Copilot in ExcelAgentic edits, PivotTables and charts in ExcelExcel on Windows, Mac, web, mobile$9.99 a month (Microsoft 365 Personal)Standard access for business users, fair-use limits
Claude for ExcelCell-cited edits and models in open workbooksExcel add-in on web, Windows and MacClaude Pro $20 a monthNone, needs a paid Claude plan
ChatGPT WorkMulti-step agent tasks ending in filesChatGPT desktop, web and mobilePlus $20 a month, Pro from $100Free and Go: limited, desktop app only

The tools to weigh

  1. 1. Numeric

    Numeric

    Good for
    Close management and analytics, reconciliations, and agentic flux and custom reporting that run directly against ERP data, plus cash management, ASC 606 billing and revenue for usage-based models, and a financial data platform the company positions as an ERP replacement.
    How it works
    Cloud close platform, with an ERP-replacement option
    Runs on
    Numeric does not disclose its models. Its AI surfaces through in-app Prism reporting, Agent Builder and MCP access from Claude or ChatGPT.
    Price
    Not published, quoted by transaction volume and seats.
    Choose it when
    Your team wants the close run in one governed place against live ERP data, with flux and reporting drawn from the reconciled numbers.

    Sources: Numeric homepage (products, demo request), Numeric pricing page (no prices shown), Ramp: Numeric vs FloQast comparison, September 2026

  2. 2. FloQast

    FloQast

    Good for
    Close management with Close AI agents that match transactions, automate reconciliations and post journal entries, layered onto a close process with SOX and COSO controls governance, for accounting teams on NetSuite, Sage Intacct and similar ERP suites.
    How it works
    Cloud close platform
    Runs on
    FloQast does not name its models (November 2026).
    Price
    Not published. FloQast prices on business outcomes rather than per-user fees, and no trial is advertised.
    Choose it when
    Your team already runs its close and SOX controls in FloQast and wants AI agents inside that governed platform.

    Sources: FloQast pricing page

  3. 3. Basis

    Basis

    Good for
    AI agents for accounting firms and corporate teams across four portfolios: CAS and advisory work that takes the month-end close review-ready, tax returns built to review-ready workpapers, audit testing, and corporate accounting, with agents working alongside staff and their work passing through human review.
    How it works
    Agent platform that connects to a firm’s systems
    Runs on
    Third-party OpenAI models: o3, o3-Pro, GPT-4.1 and GPT-5 in an OpenAI case study from 2025, with Basis posting about GPT-6 Astra in 2026.
    Price
    Not published, demo-led, and no trial is advertised.
    Choose it when
    Your firm needs agents working across many client books, in close, tax and audit, with staff starting at review rather than preparation.

    Sources: Basis homepage (portfolios, human review), OpenAI case study on Basis models

  4. 4. Ultramarine Work

    Paper Instruments

    Ultramarine Work suits accountants whose close lives in files: staff accountants on reconciliations, lease accountants keeping ASC 842 schedules, assistant controllers on the close package. It drafts schedules, classifications and entries from trial balance exports, subledger details and lease PDFs, citing sources and flagging what it cannot trace. If the close belongs in a governed platform, look at Numeric, FloQast or BlackLine, and at Basis for delegating across clients.

    Ultramarine Work runs Ultramarine 1 Flash, Paper Instruments’ own model, post-trained on GLM-5.3-Flash. Paper Instruments writes the modes and trains the model, so both improve together. Accounting & Close, one of over 130 professional modes, serves corporate accounting teams, with 19 skills spanning close, reconciliations, leases, intercompany, flux and tie-out review. The mode stays selected for the workstream, and the agent picks the skill each step needs. Standard runs Ultramarine 1 Flash, Assisted adds a frontier model as advisor, Advanced hands the task to a frontier model.

  5. 5. BlackLine

    BlackLine

    Good for
    Enterprise financial close and account reconciliation at multinational scale, with the Verity agent layer preparing reconciliations, matching transactions and drafting variance explanations.
    How it works
    Enterprise close and reconciliation platform
    Runs on
    BlackLine does not disclose its models.
    Price
    Not published, quote-led with implementation, and no trial is advertised.
    Choose it when
    Your group already runs BlackLine for enterprise reconciliations and wants governed agents inside the controls its auditors already review.

    Sources: BlackLine press release: Verity Prepare general availability, July 2026

  6. 6. Copilot in Excel

    Microsoft

    Good for
    AI inside Excel itself: an edit engine that makes multi-step changes across large workbooks, a plan mode that builds PivotTables, charts and conditional formatting, and chat that answers questions about the open sheet.
    How it works
    Feature and add-in within Excel
    Runs on
    Microsoft says the edit engine runs on Anthropic models, and a January 2026 Copilot post describes choosing GPT 5.2 or Claude Opus 4.5, Auto by default.
    Price
    Microsoft 365 Personal at $9.99 or Premium at $19.99, or the Copilot Business add-on at $21 a user a month, $18 on the promo running to 2026-12-31.
    Choose it when
    The deliverable is an open workbook and your team already standardizes on Microsoft 365, so the help arrives inside the file.

    Sources: Microsoft support: Copilot in Excel modes and edit engine, Microsoft 365 Copilot pricing

  7. 7. Claude for Excel

    Anthropic

    Good for
    Claude working inside Excel: it edits models with cell-level citations, updates assumptions while preserving formulas, builds and populates whole models, and pulls market data through S&P Global, LSEG and Daloopa connectors, drawing usage from the Claude account you already have.
    How it works
    Excel add-in, part of Claude for Microsoft 365
    Runs on
    A curated subset of Anthropic Claude models, optionally routed through Amazon Bedrock, Google Vertex AI, Azure AI Foundry or a gateway.
    Price
    Needs a paid Claude plan: Pro $20 a month ($17 billed annually), Max $100, Ultra $200. Not on the free plan.
    Choose it when
    The workbook is the deliverable, your team already pays for Claude, and cell-level citations should make each edit checkable.

    Sources: Anthropic docs: Claude for Excel, Anthropic pricing

  8. 8. ChatGPT Work

    OpenAI

    Good for
    OpenAI’s agent for long, multi-step jobs that end in finished files: it builds spreadsheets, decks and documents across connected apps, and runs in ChatGPT on the web, mobile and desktop, with Free and Go users limited to the desktop app.
    How it works
    General agent inside ChatGPT
    Runs on
    OpenAI’s own models: GPT-5.6 at launch (July 2026), with GPT-6.1 Sol rolling out from September 2026.
    Price
    Free and Go get limited Work on desktop only, then Plus $20, Pro from $100, Business $25 a month ($20 annual, two or more users), with Enterprise quoted.
    Choose it when
    One task should reach across connected apps and end in finished files, and your team already works in ChatGPT.

    Sources: OpenAI ChatGPT pricing, OpenAI help: ChatGPT Work and Codex

Pricing and how to try it

Plans start at $5 a month, and Starter and Plus begin with 7 days free. On our internal PI Knowledge Work benchmark, Ultramarine 1 Flash costs about $0.13 a task, about a tenth of Claude Opus 5.5 or GPT-6 Sol at comparable scores.

Start free trial See every plan

Questions

Is this a system of record for the close?

Ultramarine Work is not a system of record. It builds and checks files from your exports and connectors, and the ledger, the ERP and the close controls stay where they are, which is why teams running Numeric, FloQast or BlackLine can add it without a migration.

Will it post journal entries or file anything?

It does not post journal entries, and it does not file, sign or send anything. Work prepares schedules, reconciliations and drafted entries, a person decides what is used and posted, and that holds in every way of working, including Advanced.

Which models do these tools use?

Model disclosure varies across the eight. Claude for Excel runs Anthropic’s Claude models, Microsoft says Anthropic models power Copilot in Excel’s edit engine, and ChatGPT Work runs OpenAI’s own models: GPT-5.6 at launch (July 2026), with GPT-6.1 Sol rolling out from September 2026. Basis has used OpenAI models, and Numeric, FloQast and BlackLine name none.

Are these tools free to try?

A few have limited free plans. ChatGPT Free and Go include limited Work on the desktop app, Claude for Excel needs a paid Claude plan from Pro at $20, and Numeric does not publish its price, while FloQast, Basis and BlackLine quote only. Ultramarine Work has no free tier, but its Starter and Plus plans begin with 7 days free, then cost $5 and $20 a month.

Will AI output pass our auditors?

Only what a reviewer can tie to support passes as evidence. Figures need to trace to their source files, models are named and dated, and someone signs before release. FloQast and BlackLine reduce the work of SOX evidence inside their controls platforms, and an untraceable figure does not go in the file, whatever drafted it.

Does Accounting & Close replace our review controls?

No, the review controls remain yours. Accounting & Close drafts against the mode’s review standards, segregation of duties, approval and sign-off stay with the team, and the agent never posts, files or signs anything on a reviewer’s behalf.

Sources

Checked on